Industrial paper production and the world of artificial intelligence seem poles apart. But advances in both industries have opened opportunities for forward thinking manufacturers to create more efficient, flexible and greener businesses, better able to serve changing markets and client demands and building on earlier data initiatives in PiD and Model Predictive Control.

A global move from single source plastic to paper, a move from virgin pulp to recycled and the rise of home delivery shopping have led to increased demand from flexible, low-cost, high- throughput operators in this $1 trillion industry. Consultancy company McKinsey estimate efficiency savings through digitisation alone will be $20bn by 2025. At Curious AI, we see this as simply the start, so what are the current challenges and how can AI address them?

The modern industrial Pulp & Paper mill is a huge, capital intensive facility, where complex chemical processes consume energy and water and produce chemical waste and effluent as by-products. The industry needs to keep machinery operating at maximum capacity, whilst minimising the possibility of process failures. Finished products need to be of consistent quality, and the plant needs to retain as much energy and re-use as many resources as possible.

These problems are characterised as complex control and optimisation problems and for the first time they can be solved through advances in Reinforcement Learning – goal oriented algorithms that can develop super-human performance over time. By pairing these algorithms with Deep Neural Networks, we have a system that can build a model of complex processes and then learn to control and optimise them through learning, rather than having to explicitly program in every possible scenario, impossible in something as complex as a paper mill.

At Curious AI, we call this “Neural Model Predictive Control” a digital engineer, building a model of the entire process, data-point by data-point, then simulating and evaluating every possible action to maximise a goal. The scenarios are evaluated by our neural network, which is always thinking about our overall objectives (running at maximum throughput, keeping the evaporation tanks working at optimum temperature, reducing the threat of roll-breaks, etc) and our reinforcement learning algorithms take the best immediate next action to optimise the goal.

The benefits are tangible. Capital utilisation is increased as the AI models and optimises reel speeds, whilst simultaneously creating probabilistic models that avoid reel-breaks or production quality failures. Material efficiency is increased, with model control of chemical blending and fibre mixes, particularly important where recycled assets are added to the process.

Energy efficiencies are achieved as the model smooths the demand for steam and general energy to near-term and medium-term requirements. Finally, the ability to meet ecological targets is improved with better energy and material recovery. Eventually, we believe the technology will allow mills to be redesigned giving even greater benefits.

Curious AI have developed the world’s only Neural control technology, capable of controlling and optimising complex industrial systems in real-time, allowing paper mills to increase tonnage whilst simultaneously reducing costs. So, whether you are a full-process mill, or a niche manufacturer, we can help you on your journey to an AI driven future.

Write a Comment

Your email address will not be published. Required fields are marked *