AshCycle – AI-Assisted Industrial-Urban Symbiosis

By Juhani Teeriniemi, Lead Developer at Ai4Value Oy

I have recently been involved with the AshCycle Project, which is led by University of Oulu. It aims to reduce waste by improving the utilization rate of ashes from incinerators. AI4Value‘s task is to develop a digital tool that makes the results of this project accessible for a wider audience.

For me, it has been like return to my roots.  Who would have guessed that I will need to recap course materials from “Chemistry of ashes of solid fuels” by Minna Tiainen, a course I took at University of Oulu in 2011. I must admit that the folder was somewhat dusty.

Our digital tool consists of two main components: application assessment and life-cycle assessment. The former matches ash properties to suitable applications, while the latter evaluates both monetary costs and carbon footprint.

On the most basic level, the application assessment is based on a set of requirements for each application. For instance, if you want to use ashes in agriculture, you must comply with certain limits for concentrations of toxic elements (such as mercury, lead, cadmium). This will rule out some ashes. Then there is the question of cost-efficiency and finding the best delivery configuration between complex network of operators.

Machine learning and artificial intelligence (ML & AI) can help in several ways. It can provide estimates about application feasibility even when the user has not measured all required properties for a new ash material. The solution can also learn more refined combinations of requirements by getting feedback about the end-products. Logistics optimization is a key part of life-cycle assessment. What are the supply and demand capacities? Which operators should we use? Is it cheaper to just dispose everything to the nearest landfill?

Sometimes the results can be surprising. For example, the tool was able to predict a chemical compound by using concentration of another unrelated compound. This seemed like some kind of bug. Is the system cheating? From where does it get extra information? Then we realized that it was able to infer source of the material and that linked the concentrations together. This kind of undirect associations can be very difficult for human to notice. Especially if several features involved.

I am excited to see what kind of new ML/AI tools will spawn from this project. No doubt these tools will be applicable to many other applications that can abstracted to fit into to this framework.