Today
Raw values do not show how well a reverse osmosis plant runs: when the water gets colder in winter, the plant delivers noticeably less permeate although the membranes are clean. So cleaning follows the calendar or gut feeling, sometimes too early, sometimes too late. Early fouling or scaling only shows once cleaning no longer helps and membranes have to be replaced.
An example
Aldevo Wassertechnik looks after a two-stage reverse osmosis plant with 50 m³/h of permeate at a beverage maker. Since October, permeate flow has dropped by a quarter, and the operator asks for a cleaning. The agent normalises for the water temperature, which fell from 19 to 11 °C: normalised, only 3% is missing, so the membranes are not blocked. What stands out instead is the normalised differential pressure of the first stage, up 16% in three weeks, together with an SDI of 4.2 in the latest lab report. The agent proposes an alkaline clean of the first stage only in week 49 and new cartridge filters; the service technician checks the trends and agrees the date with the operator.
Your data, wherever it lives
The operating data of your reverse osmosis plants sits today in controller exports and operator logbooks, lab values in PDF reports, cleanings and membrane changes in the maintenance system, and operators' weekly readings in e-mails and spreadsheets. demi reads, understands and links them per plant and stage, with your rights. AI normalises every reading at once and shows when a cleaning is due; when to clean is decided by your service technician with the operator.
Where the data lives today
- Machines & sensors
- Databases
- Office files
- PDFs
- E-mails
- Scans & paper
How demi solves it
Three layers in this order: connect the data first, then the rights, then the AI.
- 1 · ConnectData Layer
The case needs data from the plant controller, the lab, the maintenance system and spreadsheets. demi replaces nothing and links, per plant and stage, pressures, flows, conductivities and temperatures with lab values such as SDI, hardness and silica, with antiscalant dosing and with every cleaning, its chemicals and its result. demi only reads the controller and never acts on the plant.
- 2 · PermitPermission Layer
Service sees the plants of the customers it looks after, each operator only its own. Cleaning proposals reach the operator only after release by the responsible service technician, and every cleaning is recorded in the plant file with date, chemicals and name.
- 3 · ActIntelligence Layer
The agent normalises permeate flow, salt passage and differential pressure to reference conditions following ASTM D4516 and compares them with the state after the last cleaning. From the pattern it tells whether biofouling in the first stage or scaling in the last stage is more likely, and proposes the cleaning, chemicals and date. It computes the values with fixed formulas; the language model only explains.
How it runs
- 1
Operating data arrives daily as an export from the controller, lab values as reports and the operators' weekly readings by e-mail.
- 2
demi normalises permeate flow, salt passage and differential pressure per stage to reference temperature, pressure and salinity.
- 3
The agent compares the normalised values with the state after the last cleaning, with the lab values and with the dosing.
- 4
When a value crosses its threshold, it proposes cleaning, chemicals, stage and date, with the trends as evidence.
- 5
The service technician checks and agrees the date with the operator; the cleaning and its result go into the plant file.
What changes
Cleaning when the membrane needs it, not by the calendar
Fouling and scaling caught while cleaning still works
Longer membrane life, less chemical use and downtime
What does it bring you?
Check it with your own numbers
Every starting value is an assumption you can change. The calculation sits right below, step by step.
Your numbers
Result
+192 hfreed per year
Assumptions and calculation
demi normalises the operating data and shows when a clean is due; you plan it with the operator.
60 min → 36 min per case
How we calculate
40 plant reviews × 60 min × 40% × 1.0 (manual) × 12 months = 11,520 min = 192 h
Value per year, before demi costs
Freed per year+192 h
Assumption: labour cost of €49.50 per hour in German manufacturing in 2025 (Destatis), plus workplace, IT and overheads.
= Value per year, before demi costs
+€12,480
gained per year
192 h ÷ 1,760 h per role = 0.1 full-time roles
An estimate, not a promise. The starting values are assumptions for an industrial company with 150 to 500 people.
Combine with more cases and get the packageQuestions about this use case
Why is permeate flow alone not enough?
Because flow swings with water temperature, feed pressure and salinity. With every degree of colder water, a membrane delivers noticeably less permeate without being fouled. Only normalising to fixed reference conditions separates that swing from real fouling or scaling.
When does the agent propose a membrane clean?
When a normalised value crosses a threshold you set or take from the membrane manufacturer: less flow, more salt passage or more differential pressure than after the last cleaning. Each proposal shows which value triggered it and how fast it is changing. Whether and when to clean is decided by the service technician with the operator.
Can demi tell biofouling from scaling?
The agent reads it from the pattern and the lab values: if differential pressure rises mainly in the first stage, that points to biofilm or particles; if flow drops in the last stage while salt passage rises, it points more to carbonate, sulphate or silica scale. It names the matching cleaning, alkaline or acid. When the data is not clear, it says so; a membrane autopsy then settles it.
What if the RO plant has no interface?
Then controller exports as files, operator logbooks as spreadsheets or photos and lab reports as PDFs are enough. How the data gets to demi, we clarify in conversation. demi only reads and never writes back to the controller.
