Due to the role I hold at ABB and based on my earlier experience with Big Data analytics and digital development at companies, such as eBay, Oracle and Rolta, industrial automation professionals frequently ask me to explain what Artificial Intelligence (AI) can do to help them make the next big leap in extracting value from digitalisation. An excellent way to explain this is via an analogy with a Rubik’s Cube, the famous 3D combination puzzle that offers interesting comparisons to industrial AI. Imagine holding a Rubix cube; most of us have solved or perhaps tried solving this intricate puzzle. Depending on the skills, it can take just a few minutes to solve or one might go on for days and not make progress.
But if one has three additional items, he/she would likely be much quicker at solving it. These are:
A holistic and/or cross-dimensional 360-degree view.
Proven algorithms based on different patterns.
Greater speed to manage feedback, corrections & iterations.
In the industry, the same ‘quicker resolution’ scenario is applicable to optimise operations and assets since, just like the Rubik’s Cube, complexity in industrial operations is multi-dimensional, multi-variant and dynamic.
The recent advent of technologies, computing and networking innovations, such as Industrial Internet of Things (IIoT), is fuelling new ways to address industrial complexities and better use the power of data. This presents a new paradigm of ‘industrial AI’, thus bringing a higher degree of prediction accuracy & optimisation to operations, processes and assets. This is allowing us to truly apply AI to industries, by combining it with specific industrial domain expertise (for example in cement, chemicals, food & beverage, marine, metals, mining, oil & gas, pulp & paper, refining, etc) for safer, smarter, more efficient and sustainable industrial operations.
Ensembled approach, instead of discrete silos
The real solution for successfully applying industrial AI comes by using what I call an ‘ensembled approach’. This is not unlike the idea that the ancient Greek philosopher, Aristotle, had around 2400 years ago, when he stated, “The whole is greater than the sum of its parts.” When individual parts are connected, they are worth more than if the parts were in silos.
The same is true with data silos. Applying AI to industries, using the potent combination of specific industrial domain expertise, contextual data, advanced technologies and AI algorithms, one can extract far greater value than the individual silos would ever be able to generate on their own. This can result in safer, smarter and more sustainable industrial operations that are more productive and profitable. This reality is before us today.
Why industries need AI
In today’s highly competitive business landscape, industries must find every possible operational excellence advantage that enables them to produce the highest quality products, at the fastest rate, with the best and most reliable asset performance and cost-effectively. With production machinery being pressed to maximum capacities, much of which utilises state-of-the-art industrial automation to enforce efficiencies on the process, the key to achieving additional productivity gains is by extracting value from the data being generated by and stored in existing operational, engineering and Information Technology (IT). But many companies struggle to know where to begin this digitalisation journey.
Today, according to industry experts, such as the ARC Advisory Group, the average industrial plant uses less than 20% of the data it generates. Successful companies will be those that can capitalise on data to a much higher extent to turn it into actionable insights at the right time, thus helping them efficiently improve performance and optimise costs.
The ‘X factor’ to contextualise OT, ET and IT data
In our view, the foundation of industrial AI is ‘contextualised data’, because its value exponentially increases when one contextually integrates Operational, Engineering and Information technology data (OT, ET and IT). We call this the ‘X factor’.
If we take a simple example of a critical asset, such as a turbine, the OT data tells us how this asset is performing along with the health indicators. ET data tells us how well or poorly it is performing compared to its specifications and how compliant it is regarding the integrity and safety of the operations. IT data indicates how to keep the asset performing at the desired condition with the right maintenance strategy, spare parts inventories, risk & investment planning, etc.
Obviously, asset failures not only for a turbine but even for smaller systems or components, like motors, heat exchangers, compressors, pumps and bearings, can have a big impact on production and profitability. Many failures can be avoided, as analysis from ARC illustrates -
Only 18% of the asset failures, on average, show age-related failure patterns
But 82% of the asset failures occur mostly due to the impact of the related assets, process parameters, lack of adhering to critical asset management and reliability practices, ignoring key recommendations, etc
The above findings demand that various asset data and information be analysed as a contextualised system for prediction and optimisation.
When contextualised data is applied using industrial analytics, we can then gain actionable insights that support better strategic and tactical decision making. Then, by applying industrial AI, domain expertise and Machine Learning (ML), one can develop true predictive and prescriptive analytics for optimisation that will directly drive better business outcomes.
These advances now allow us to bring greater value to a wider spectrum of roles, based on the same intelligence sources. This provides deep, cross-functional and actionable insights for key plant and enterprise decision-makers, including operations, maintenance, reliability, engineering, IT, digital, financial, capital planning, etc.
How industrial AI has evolved
Historically, OEMs provided models based on first principles, meaning basic physics, and they were deemed good enough for an isolated physical phenomenon or piece of equipment. Later, industrial automation providers and process owners extended that to predict and address behaviours through control system engineering and simulation modelling.
However, manufacturing and other industrial domains are complex and nonlinear by nature, making it impossible to use first-principle physics models. Simple ML models, while better, only serve to the extent for which the dataset has been trained, which doesn’t cover all conditions. This is where industrial AI comes in.
Large benefits become easier to harvest
The rewards of adopting industrial AI, on a large scale, can be significant, with some analysts, such as Accenture Research and Frontier Economics, predicting productivity gains as high as 40%. The consulting group McKinsey estimates that manufacturing digitalisation and industrial AI could boost EBITA by three to five points. McKinsey says IIoT-based process optimisation could lead to about 9% throughput increase and IIoT-based predictive maintenance can result in a 10-30% reduction in maintenance spend. But it should be noted that these figures are based on making the new technologies work on a large scale in a plant or enterprise.
To reap such rewards, ideally, a comprehensive industrial AI suite should:
Seamlessly integrate data from heterogeneous source systems, treat the data for its quality and data integrity and extract its patterns through feature engineering models
Provide intuitive, user-friendly model development and training capabilities
Provide powerful predictive and simulation engines with pre-built and extensible optimisation models
Offer role-based visualisation
Integrate rapidly deployable analytics business value applications
AI = Higher ROI
The key takeaway here is that industrial AI has reached a stage where it can easily be adopted and be used to unlock new values. Especially for asset-heavy sectors, like process and energy industries, transportation and other industrial applications, this gives significant benefits in predictive analysis and optimisation of key assets and processes to boost overall ROI.
With industrial AI reaching a tipping point through new solutions, such as ABB’s industrial analytics and AI suite, automation is taking another leap towards enabling more autonomous operations. So, I would say that now is a good time for companies to start or accelerate their data utilisation journey with these new advanced solutions.