By Wilhelm Swart, MD: 4 Sight Operational Technologies
There’s a buyer’s market in AI and machine learning (ML) at present, with every indication that more spend is being focused in that direction. Nevertheless, around 28% percent of AI projects have failed, and many others fail to make a significant impact.
Part of the problem is that these projects often take place in isolation, are developed by the company, and have a weak business case. Skills shortages, lack of clean data and other challenges limit the usefulness of these projects, preventing them from being integrated into how the business runs.
Industrial AI offers a different, much more promising approach. Industrial AI creates a systematic, collaborative way of solving specific industrial challenges by embedding algorithms into applications used by the business. A key element of industrial AI is that the algorithms are based on deep domain expertise.
It’s worth unpacking two important concepts to understand industrial AI more deeply. The first is that industrial AI looks beyond single-use, standalone projects to create a platform on which the AI can be scaled and developed as business needs change, and then systematically rolled out to other areas of the business with minimal risk.
This systematic, platform-based approach is critical because AI requires significant groundwork in terms of creating data-collection strategies across the entire organisation to ensure consistent—and positive—results. Imperfect or scanty data compromises the value of AI from the get-go.
The second is the need for deep domain expertise. All too often, AI is treated as a project undertaken by a skilled data scientist, but in truth it is very difficult to understand which data is important, and to distinguish between mere correlations and actual causation. Only those with a deep domain understanding are in a position to identify which apparent “insights” are applicable to the business and can genuinely improve it.
Taking a step back, one could argue that what really distinguishes industrial AI is the way in which it is embedded into the business, drawing insights from the data its systems generate and then feeding improvements back into the business. In this way, industrial AI generates benefits that are tangible, with the resulting savings or improvements creating the space to fund further projects.
Compelling use cases
Mining and manufacturing are highly capital-intensive industries, with much of that capital tied up in plant and equipment. They are also very complex, so the scheduling of all the various business processes is also critical to meeting client commitments and maximising the return on capital. In such an environment, equipment downtime has a massive knock-on effect, resulting in lost sales, reduced cash flow and impaired brand reputation. It can also impact safety and reduce quality as corners are cut in order to make up lost time.
It’s estimated that Fortune Global 500 manufacturing and industrial firms lose 3.3 million hours a year to unplanned downtime, the equivalent of 8% of their annual revenues. The figure is similar in mining (1.2 million unplanned downtime hours a year).
Industrial AI can make a significant impact here by using both historical data and data from equipment sensors to develop robust abilities to predict breakdowns so that preventative maintenance can be scheduled. More reliable equipment makes the entire complex process of running a plant, a mine or any other complex industry environment much easier, with immense benefits in terms of brand reputation, competitiveness, profitability, and worker safety and engagement.
Predictive or preventative maintenance accounted for a quarter of all the industrial AI implementations in 2019 because of the immediate and quantifiable benefits it delivers. From our own experience with clients, industrial AI can improve plant availability by up to 25% and reduce maintenance costs by 20%.
A second, related benefit is the way that industrial AI allows companies to shift their focus to quality and reliability, rather than simply throughput. The aim here is to understand the impact of all the variables within a complex process, and then control them to get a better, more consistent yield. An example here could be the ability to adjust better to variable feedstock in steelmaking, to achieve a consistent result.
The ability to maintain quality, reliably and sustainably not only saves money, it supports profitable growth and brand reputation.
A third use case relates to the use of embedded AI and ML to optimise processes in real time. For example, we are reliably able to help clients improve their energy efficiency and correspondingly reduce their carbon footprint by 15-20%. Other gains we have helped clients achieve include a 3-4% increase in yield via increased plant productivity.
Industrial AI can deliver significant benefits in the short term, as I have argued. But perhaps even more exciting is its potential to gradually enable complex plants and industrial processes to become self-learning, to automatically learn from experience to become more efficient at delivering better quality product reliably.
Now that is exciting.