OT-focused AI Systems Accelerate Adoption of AI in Manufacturing

By Craig Resnick

Category:
ARCView

Summary

New digital technologies have the potential to augment people and processes to an unprecedented degree across the process manufacturing industries. New commoditized computing resources in the edge, fog, or cloud; combined with artificial intelligence (AI) are changing the way people work and how companies perform.  Process manufacturers strive to be recognized as dynamic and empowering companies.  New, AI-enabled solutions can accelerate adoption of AI in aicr1.JPGmanufacturing and help them achieve that goal.  AI is a major component of digital transformation, which touches industrial products, operations, value chains, and aftermarket services.  This transformation is enabled by augmenting people and knowledge through expanded use of sensors, data, and analytics.    

While AI is currently being used for advanced data analytics and to generate new insights predictions, prognostics, or recommendations, ARC Advisory Group does not yet see many process manufacturers using AI to impact their conversion costs or production efficiency in any significant way.  Until this happens, AI by itself cannot be considered transformational.  Ultimately though, companies will make AI part of their advanced control strategies to help close the loop between analytics and action.

Making this transition requires making AI and machine learning part of the toolset used by process and automation engineers; it requires democratization of AI with the process manufacturing subject matter experts (SMEs).

Tackling the Data Science Talent Shortage

AI Becomes Useful When Combined with Deep Domain Expertise of the Application and Problem  aicr2.JPGIT-OT convergence is a much-discussed topic relative to digital transformation.  A main benefit is to make enterprise data systems ready for analytics and AI.  Transformational value from this data is extracted by applying AI technologies, such as machine learning. AI becomes useful only when combined with deep domain expertise of the application and problem. However, given the tools and technologies available for AI implementation, most of the AI work can only be performed by data scientists.  Some manufacturers have seen increased success by creating focused teams made up of data scientists and SMEs, but this is neither practical nor sustainable for most manufacturers, especially for the medium and small manufacturers that can least afford to lag behind in digital transformation.  

 

Democratization of AI for Manufacturing SMEs

The key change that is making AI pervasive and useful is the “verticalization” of AI for specific businesses and processes. One way to achieve verticalization is by creating specific packaged applications such as those being built for finance, insurance, or e-commerce. However, sustainable verticalization of AI also requires that applications be scalable, which may not yet be feasible through packaged solutions in many business verticals. A viable alternative for such verticals is democratization – letting SME’s create and sustain AI solutions at scale.  ARC Advisory Group believes manufacturing is one vertical in which scalable applications are feasible with democratization. While similar equipment and processes are used in many cases, each plant and application has its own specific nuances, challenges, and dynamics that the SMEs that run these plants understand.  Democratization is often the best alternative for widespread and sustained adoption since it makes AI just another solution in the toolset of manufacturing SMEs.

As ARC learned in a recent briefing, Quartic.ai, a young and forward-thinking company, has developed a platform designed to make it possible to democratize AI with process manufacturing SMEs.  The company provided some examples to support this claim.

AI Platforms Must Focus on the OT Users

To democratize AI for manufacturing SMEs, platforms and systems must have some key attributes. First, to reduce the need for extensive knowledge of data science, statistics, and algorithms; machine learning must be automated and extracted.  Second, representation and user experience must be in the language and techniques that most process engineers, equipment reliability, and automation professionals understand and use every day. Third, the intelligence created by AI must integrate seamlessly with existing OT systems and existing work processes so as not to be disruptive and bring on additional change management challenges.

Building Trust in AI Accelerates Adoption

AI in manufacturing aicr3.JPGWhile automation and abstraction of the machine learning process helps users focus on the problem and subject matter expertise, using a black-box approach to machine learning can lead to a lack of trust and hinder adoption. Quartic.ai is addressing this with a “human-in-the-loop” approach. While many machine learning tasks such as variable relationship extraction and algorithm selection are automated, the user selects the final model. The user is also allowed to train multiple models using different combinations of features and can verify the model’s output behavior by changing the values of the features. Feature engineering is perhaps the most valuable and challenging part of machine learning. By leveraging the subject matter expertise of the user, better models can be built and model training time and valuable compute resources optimized.

Quartic.ai’s Platform Built for Industrial Users/Environments

Unlike the many other AI solutions designed for use by data scientists, Quartic.ai built its platform to recognize that OT specialists and subject matter experts in industrial plants also need these types of tools.  The company also understands that rapid deployment and fast ROI are often prerequisites for launching new solutions in industrial environments and that data science expertise, in general, is in short supply.  Furthermore, the platform conforms to ISA99 security standards, enabling secure data connectivity between legacy and modern assets.

The Quartic.ai platform is designed to help accelerate Industrial IoT and AI deployment by eliminating many current barriers.  This can reduce both time to deployment and associated costs.  The platform makes use of the process manufacturer’s own data to provide new and empowering insights for industrial workforces.

Accelerating IIoT and AI Solutions

According to the company, the Quartic.ai platform provides a complete system for developing AI-enabled smart manufacturing, Industrial IoT, and Industry 4.0 solutions. By leveraging ML, it enables users to modernize their existing OT AI in manufacturing aicr4.JPGfor digital manufacturing or build a new smart OT system. The platform is built on two main components: illuminator and eXponence, each with a set of application modules. The platform can be deployed either on-premise or in a hybrid edge-fog-cloud architecture.

illuminator, an OT data lake, provides IIoT, OT, MES, and ERP data in dynamic context to intelligence applications in eXponence. For expert users and data scientists, illuminator provides an integrated contextual data pipe to build AI applications using a choice of tools and libraries. eXponence uses automated ML, a rules engine, and complex event processing (CEP) to build the intelligence, provide visualization, and communicate the intelligence. Visualization and communication functions can alternatively be performed with the legacy or new OT system by connecting the intelligence output from eXponence.

While built using AI techniques and technology, the eXponence intelligence engine is intended for use by OT professionals and manufacturing SMEs. The Quartic.ai platform allows users to build IIoT systems in accordance with the Industrial Internet Reference Architecture (IIRA, ISO/IEC/ IEEE42010:2011)1 and RAMI 4.02.

illuminator IIoT Data Engine

The illuminator IIoT Data Engine consists of several key components.  Whether for a single process unit site or the entire enterprise, Quartics AssetHarbor provides data abstraction and context, enabling continuous extraction of context from multiple data streams.  Process operations, both batch and continuous, are abstracted in the context of an asset. The illuminator data pipe then allows publish-subscribe access to the attributes of asset objects for building smart applications. Intelligence and insights created using Quartic’s MetaTrainer, Reckon, or external applications are added to the asset objects using extensible attributes. The reference data models can be custom, asset-oriented standards, such as ISO14224; or manufacturing oriented, such as ISA95.

AI in manufacturing aicr5.JPGOther key illuminator components are OT Data Lake and Data Pipe. illuminator ingests data streams from multiple sources and protocols for a common reference. This low-latency, high availability publish-subscribe data pipe provides access to real-time abstracted data to build AI applications with Quartic’s eXponence and serves as an OT data lake. Advanced users and data scientists can connect with illuminator for contextual abstract data types to be able to use ML tools and libraries of their choice and eliminate the need to integrate data to speed deployment. Connectivity with illuminator is possible using Kafka Pub-Sub, Illuminator API, or SDK.

Quartic’s data integration architecture, Qnnect, is a hardware-agnostic, distributed edge and fog software system for IIoT. Edge Gateways (Qlite) connect to sensor networks, PLCs, PACs, and other edge devices using legacy and modern protocols. Fog nodes (QPro) connect to multiple QLites to create a scalable, layered architecture aligned at each ISA95 level. QRelays may be used between levels to comply with ISA99/IEC62443 zones. Endpoint security and data encryption provide an additional layer of security. Distributed storage and compute allow ML to be executed on fog and edge nodes.

eXponence Intelligence Engine

The eXponence Intelligence Engine consists of several application modules. ContexAlyze discovers and analyzes patterns in the data. Understanding data in the context of operations and asset condition behavior is a critical step to define, design, and develop ML applications. The visual analytics in ContexAlyze allow OT users to apply their expertise to help discover patterns and features to build and validate hypotheses for ML design. Users can perform multi-variable pattern searches, AI in manufacturing aicr6.JPGcreate and store event frames, and perform cohort analysis for batch processes and “golden batch” analysis. Users can also collaborate on data discovery with annotations and by sharing snippets.

MetaTrainer, another key eXponence application module, is an automated ML tool for industrial subject matter experts.  MetaTrainer allows the user to focus on his or her expertise to build AI applications without requiring coding or programming.  ML generates feature significance, but users can select features based on their domain knowledge. Multiple models are trained and scored in parallel, recommending the best model to be deployed on real-time data.  Visual techniques for interpreting the model outputs provide understanding, which can further increase trust in and adoption of AI.

The intelligence engine also features Reckon logic and calculations to build insights, KPIs, and alerts on both real-time process and condition data and the output of ML algorithms.  This visual, user-friendly interface allows users to create simple logic and calculations or advanced equations without requiring scripting or coding. Quartic View, an insight visualization tool, is used to present equipment health and operational performance KPIs. Widgets support visualization of workflow status, actions from insights and alerts generated by ML and logic.

Conclusion

In today’s increasingly complex global competitive environment, applying Quartic.ai’s vision of human-in-the-loop AI for subject matter experts could help process manufacturers make decisions that enhance conversion, improve production AI in manufacturing aicr7.JPGefficiency, and optimize their human and physical assets. This means being prepared to embrace digital transformation to acquire the needed information, even if it alters current processes.  It is also particularly important to select partners, such as Quartic.ai, that have expertise in process industry applications and the related software platforms and understand the way subject matter experts in these industries apply technology for process control and asset reliability. 

But technology alone is not the solution. It’s critical to also know how existing personnel can use the tools to predict future outcomes and make better decisions without requiring the skills of a data scientist. Technologies, such as on-line machine learning, can provide process manufacturers with the competitive edge that, ultimately, will deter-mine who wins and who loses.   Only process manufacturing companies that can manage their cost structures effectively will survive long-term.  Effective digital transformation can mitigate this risk to a significant degree.

ARC Advisory Group clients can view the complete report at ARC Client Portal 

If you would like to buy this report or obtain information about how to become a client, please Contact Us     

Keywords: Artificial Intelligence (AI), Digital Transformation, Quartic.ai, Machine Learning (ML), Data Analytics, IIoT, APC, MPC, IT-OT Convergence, SME, ARC Advisory Group.

 

Engage with ARC Advisory Group