Three kinds of artificial intelligence: from predictive to prescriptive and beyond

The use of artificial intelligence is no longer an option in many industries, but often a requirement to keep up with the competition.

State-of-the-art artificial intelligence technologies improve industrial processes, proactively detect and solve problems, and provide guidance for risk-based decisions. In doing so, they provide companies with significant cost savings and improved competitiveness.

Over the past 20 years, AI has significantly transformed the industrial workplace. Industry is deploying AI on-premises, in the cloud, at the edge, and through many types of hybrid architectures.

The three Ps of industrial AI

AI itself is not one thing but comprises a number of technologies, including neural networks, deep learning (a type of neural network), natural language processing, computer vision, unsupervised machine learning, supervised machine learning, reinforcement learning, transfer learning and others. Industry applies these types of AI in a variety of ways to create targeted solutions.

We can categorise industrial AI analytics into three different types — the three Ps of industrial AI:

1. Predictive: Predictive AI predicts when subtle changes in performance indicate bigger problems in the future. It detects incipient problems and inefficiencies, as well as errors in the design process, by analysing industrial ‘big data’ from sensors, data lakes, data historians, calculated values, audio, video, and other sources.
2. Prescriptive: Once predictive and performance analytics identify issues, prescriptive analytics provide root-cause analysis, planning and decision support, and courses of action that are most probable to remedy and optimise a situation.
3. Prognostic: This type of AI uses neural nets, deep learning, and reinforcement learning technologies to forecast future events. Industries use it to optimise monitoring, control and scheduling and to help determine how long an anomalous asset or process can continue to safely operate before it fails or loses significant function.

Predictive AI

The first of these three Ps — predictive AI — is one of the more common advanced technologies used in industry today. Although referred to as ‘predictive’, it is actually a very effective method of anomaly detection in near real time. It uses advanced pattern recognition to capture the digital signature of the normal behaviour of an asset or process. It then compares this normal digital signature with incoming, real-time data from SCADA and other control systems. It can detect deviations from the normal behaviour days, weeks, and even months before a traditional SCADA or control system alarm would trigger. That advanced notice gives companies adequate time to rectify the asset or operational problem before it’s too late.

The prevalence of predictive analytics is largely due to its general applicability to huge volumes of time-series data, often referred to as industrial big data. With the advent of the Industrial Internet of Things (IIoT), the cost of sensors has greatly decreased, allowing companies to install all kinds of online meters where they never would have previously. Those meters measure and record more data in historians and data lakes, both on-premises and in the cloud.

Predictive AI uses machine learning to find patterns in this big data. It uses proprietary data clustering algorithms help suppress irrelevant fluctuations in the data (noise) so that the AI can better detect and analyse core patterns.

Today, there are two primary types of machine learning: unsupervised and supervised. Unsupervised learning automatically analyses data and systematically determines relationships within it. It identifies deviations from patterns of normal behaviour without human intervention.

With supervised machine learning, people model assets and operations by selecting relevant sensors (tags) that are statistically related. People also select periods of archived big data that represent ‘good behaviour’ so that the software can create a digital signature of proper operation. The AI then compares incoming real-time data to this digital signature and identifies deviations as possible early warnings of asset or operational degradation. It also identifies which sensors are most indicative of each anomaly so that people can better track down the root cause of the issue and correct it before it becomes a major operational problem. This results in less downtime, better product quality, reduced risk, and increased overall efficiency and profitability.

For example, one predictive AI prevented the catastrophic failure of a turbine at a power company. A turbine had been exhibiting step-changes in vibration reductions (not increases). Each time the issue came up, the manufacturer told the customer it was OK because it was a reduction in vibration, not an increase. But, in this particular situation, the vibration reductions turned out to be due to the beginning of blade separation within the turbine stages.

The system was nowhere near to triggering a control system alarm or warning. However, had it gone on, it would have resulted in a catastrophic failure that could have destroyed the turbine, caused extensive downtime (loss of power production), and a potential for significant injury to personnel. Conservative estimates by the customer showed that the early warning detection of this issue avoided costs over US$34 million.

While companies like these have been using predictive analytics for some time, the food and beverage industry, as a whole, is just beginning to adopt these technologies. Although less mature in predictive maintenance than other industries, it is quickly finding value by monitoring and analysing production lines, reducing downtime, and improving quality.

In the food and beverage industry, predictive AI can catch irregular motor operation, where the electric current runs too high in relation to other monitored values, but not high enough to cause an operational warning. Gas oxidiser issues are another useful area for this technology, as well as conveyor problems with over-tensioned belts, and pumps running hot due to oil and valving issues.

Historically, these types of machine learning successes were difficult for novices to achieve, sometimes requiring users to write scripts and manage software code. However, machine learning software has become much easier to use, with advanced drag-and-drop graphical user interfaces (GUIs) and simple, easy-to-understand on-screen representations of identified anomalies.

Prescriptive AI

Prescriptive AI goes beyond predictive AI to recommend specific actions that operations and maintenance personnel should take to rectify an issue. For example, in the food and beverage industry, prescriptive guidance for oil issues verifies the calibration of oil temperature sensors, and then recommends replacing or recalibrating them as required. It might also recommend oil analysis to check for contamination, and then recommend replacement of the oil and oil filter depending on the results. Other types of prescriptive actions can quickly become much more complicated and involved.

Prescriptive analytics began by replacing calendar-based maintenance with condition-based triggers to create a proactive maintenance program. Now it’s critical to both improved asset maintenance and enhanced operational efficiency. Consequently, it has become an increasingly important aspect of an overall reliability-centred maintenance (RCM) program.

Prescriptive AI is revolutionising the way people work. It enhances workforce productivity and improves safety, reliability, quality and security. As a result, industry is improving efficiency in ways it never could before, and creating new types of jobs. However, AI technology is only in its infancy, and it is greatly advancing each year. The future of AI is extremely exciting, and opportunities to benefit from it are virtually limitless.

Prognostic AI

A third type of AI — prognostic AI — further enhances predictive and prescriptive analytics. Prognostic AI goes one step further by forecasting future events, such as operational performance degradation or the end of an asset’s useful life.

Prognostic AI helps people answer questions such as: “Can the system make it to the next planned maintenance outage?” or “Can the asset make it to next week, or do we need to call in emergency personnel over the weekend on overtime wages to fix the problem?”

These are critical decisions that impact both risk and costs. Risk management is a key part of what AI brings to businesses, and it can significantly help improve the bottom line of industrial operations.

The future of industrial AI

As AI continues to evolve, predictive, prescriptive and prognostic software will increasingly integrate with enterprise asset management (EAM) systems. Together, they will dynamically create work orders and use the forecasted remaining useful life of assets to prescribe actions to rectify issues. They will automate everything from issue detection, through root cause analysis, to remediation and rectification.

Beyond EAM integration, AI software will also integrate with scheduling systems to recommend the optimal time to perform emergency maintenance within the forecasted remaining useful life of an asset. Such recommendations will reduce adverse impacts on operations, minimise overall business risk and maximise profit.

AI integration will also extend to closed-loop, automated process control — in which people merely monitor fully automated and optimised end-to-end operations and maintenance processes that AI controls. These technologies exist today, and their adoption will increase over time.

Prescriptive capabilities will continue to enhance predictive analytics software so it can detect and prevent problems more quickly, better maintain industrial operations, optimise scheduling and enhance process control. Continued advancements in prescriptive capabilities will:

• better enable and empower the human workforce;
• make operations more efficient;
• improve work product and reduce mistakes;
• help people learn and transfer knowledge more quickly and completely;
• enhance safety in the workplace;
• create new jobs and business opportunities;
• ultimately, create a better quality of life.

Digital twins

A digital twin is effectively a virtual representation of a physical object or system — including larger entities such as buildings, factories, and cities. It includes IoT data, advanced computer systems, digital processes, electronic documents, and advanced analytics, which all model physical space.

AI is necessary to get the most value out of a digital twin. Combining AI with a digital twin significantly enhances productivity. This is not theory — this is a fact and is quantifiable. AI enhances workforce productivity and improves safety, reliability, quality and security.

After a digital twin is put into production, AI enhances operations for safe and profitable processes within constraints and regulatory norms. It automates monitoring and control processes through closed-loop analytics for autonomous operational control to ensure safety and performance. Many AI techniques improve maintenance by increasing the longevity and performance of assets while ensuring a safe, reliable environment for the workforce through predictive and prescriptive analytics. Various types of AI also optimise planning and scheduling by creating a self-learning approach for continuous improvement to reduce risk and maximise profitability.

In all, AI benefits industrial processes from design and engineering to operations and maintenance. It improves engineering through automated design generation, enabling lower total cost and lower risk in capital projects.

An evolving workforce

Through efficiency gains and reduced waste, AI is creating an overall greener environment with enhanced sustainability. AI also helps workers themselves. Studies show that there are not enough new qualified staff to replace the knowledge of an aging workforce rapidly approaching retirement. AI helps reduce this lost knowledge.

However, AI also disrupts jobs, which sometimes results in the elimination of certain types of occupations. This can be devastating for those impacted. But at the same, it creates a variety of new jobs, such as monitoring-service technicians, data analysts, data scientists, etc. Of course, this is nothing new. New technology has been disrupting the workforce for centuries. Ultimately, history has shown that while innovation does eliminate some jobs, it typically adds more than it destroys, resulting in a net increase in the overall workforce.

Unfortunately, AI sometimes creates a fear of the unknown, including privacy concerns and anxiety about being replaced. Companies must manage these fears and ensure that proper employee education and communication channels are in place to minimise fear due to misinformation and a lack of understanding.

Conclusion

An increasing number of industrial companies throughout the world use artificial intelligence, particularly predictive analytics. It’s no longer an option in many industries, but often a requirement to keep up with the competition.

But in order for industry workers to get the maximum benefits from AI, they need it to give them easily understood and — more importantly — actionable information. That’s why prescriptive and prognostic AI are the future of this exciting technology. By giving workers clear guidance — and factoring business considerations and risk management into its recommendations — prescriptive and prognostic AI gives industry the tools needed to maximise productivity.

*Jim Chappell oversees AVEVA’s overall AI strategy and implementation across all business sectors. In addition, he is in charge of the asset performance management (APM) suite of products and related engineering/analytics services. This encompasses industrial big data (data historians), predictive/prescriptive/prognostic analytics, business intelligence, and enterprise asset management.

Image credit: iStock.com/gorodenkoff