Teknisk Ukeblad

Digitalized old hydropower plant with cameras and machine learning

Digitalized old hydropower plant with cameras and machine learning
Userlevel 2

This article was originally published in Norwegian in TU. Read the article here


After 12 weeks, old measuring equipment could harvest data for Ringerikskraft.

The river power plant from 1978 utilizes the power in Hønefossen, after which the city is named. Neither the 26 MW Kaplan turbine nor the control system have been replaced in these 43 years, and many components are nearing the end of their service life.

Is it possible to digitize something like this? 

Yes, and it can even go fast, according to Jonas Hertel of the technology company Cognite. Within three months, they digitized analogue process values, water level gauges and alarms in the old power plant.


Uses camera instead of system data

The method is simple: Instead of extracting data via the control system or Scada, cameras have been used to read the old meters.

 -Many of the instruments in the station are off-line or only connected to a protection function. We have therefore used cameras that point down at the measuring instruments and the alarm board. It does not give the same accuracy, but in this case it was good enough, says Hertel to TU.

The cameras read both round and square meters and displays, and have so far obtained data from 68 alarms, 32 temperature signals and the drop loss over the intake gate. With machine learning, the images are transformed into digital time series. This is how you can detect changes and, at best, predict accidents.

Completely assembled in 2-3 days

Step one was to set up wifi in the old station. Within 2-3 days, cameras were also installed and online. Then it was time for fine tuning of the camera and algorithms. 

- When reading old meters with a camera, a small change in the angle of the needle can have a big impact. There was a lot of tuning in the beginning, says Hertel.

He says that both the  power sector, the oil industry and the process industry have many silo-based systems, from which it can be complicated to extract data. 

- There are many closed systems where the signals are first sent over to the operations center via scada. A lot of data is also located locally in the control system, which varies in age and often sends data in different formats, says Hertel.

Bypasses cyber security requirements 

Even if you have managed to extract the raw data and stream them to a cloud service, you will have to use a completely different method at the next facility. This meant that you also chose a camera where there was a digital water level meter you could connect to. 

- If new sensors are to be integrated in the local control system, one is suddenly inside a complex architecture, with many rules for redundancy, cyber security and more. A sensor network on the side will not affect the power plant, nor will it affect the power plant if a camera fails, Hertel explains.

In the work at Hønefoss, they extracted 20 temperature signals with an analogue to a digital converter.

- In 1978, displays were expensive, so all the temperature gauges on the control panel had one common display. Then you had to turn a reversing wheel to choose which temperature sensor you wanted to see. Now all the measurements are gathered on a PC screen. 

Reads the alarm board - and sends SMS

Another simple advance is the alarm board, designed as an old keyboard where one button lights up to indicate what is wrong. So far, the emergency guard has had to travel to the power station to check what the problem was when the alarm went off.

By pointing a camera down at the alarm board, the algorithms see which button is lit and send the guard an SMS about which alarm has gone off. 

- Now the guard can prepare already in the car. It gives security to know what is wrong, and you can call for help already on the way to the facility, says Hertel. 

Via the cameras you can look for, for example, smoke development. The system, on the other hand, is what Hertel calls "read only", and you cannot go in and control processes.


Digital water level measurement in one day

The latest major simplification is digital measurement of the water level in Damtjern, which is 17 kilometers from the power station. Ringerikskraft has had to drive up there twice a week to physically measure the water level.

Digitizing this was done in one day and at the cost of a well-stocked shopping cart at the grocery store

Digitizing this was done in one day and at the cost of a well-stocked shopping cart at the grocery store, according to Hertel.

They used a drone to take pictures and make a model. Then they set up a pressure sensor that communicates using 4G. 

- The water level sensor saves us lots of rides on steep and bad roads in places. It was a no-brainer. The sensors and cameras are not expensive, and it did not require much work either, says Anita Skagnæs, head of digitization at Ringerikskraft.

Can be used to optimize efficiency?

Although - or perhaps precisely because - the project is neither groundbreaking nor high-tech, the Hønefoss project has attracted attention when Cognite has presented it to customers around the world.

- In advance, we saw it as extremely difficult to extract data from such an old facility. We realized that it had to be done with new methods. Thus, there was machine learning and algorithms instead of connecting to the control system, says Skagnæs.

Ringerikskraft is already in the process of utilizing the data, but is also collecting them for what Skagnæs calls a treasure chest for the future .

- We sniff at how we can use the data to optimize efficiency. In Hønefoss power station, we can not do it, as the unit there is screwed together. But we also own the river power plant Vittingfoss, with five units. There you may think differently with the help of real-time data, says Skagnæs.


Inspection round with a robot dog? 


There may still be more low-hanging fruit to harvest in the old power plant. In addition to cameras and sensors, microphones are also on the list of Cognite tools. 


- There is a lot of info in the sound. You can set up simple microphones and do frequency analyzes to find out if something is different than yesterday. We have also used robots equipped with microphones to walk around oil platforms and analyze the sound image there, says Hertel. 


Cognite plans to bring the robot dog Spot to Hønefoss, to see if he can go for inspection rounds  and see that the equipment is in order. 


- Spot can carry 15 kilos with sensors, and may read data in places that are not covered by fixed cameras. Such a robot can help professionals to spend time on more interesting tasks than reading meters, says Hertel.

0 replies

Be the first to reply!