Industry’s Challenge with AI & Data Analytics – Part II

Our previous article (Part I) articulates that the adoption of AI in manufacturing industries has not yet become a widespread phenomenon despite the opposite sentiments. The main argument is that traditional industries struggle due to their incapacity to capture and manage process data properly (data integrity issues). Part II elaborates root causes in four sections.

TrueNord Global have been supporting industrial clients through their digital transformation programmes and have served a wide range of industrial clients including blue-chips and family-owned SMEs. The content on both articles intends to guide executives away from common traps along their digitalisation journeys. ?

Legacy Equipment

Depending on the specific industry, 60-80% of operations data is attributable to production processes in manufacturing. By and large, this derives from industrial process equipment used in production. Though 90% of the equipment operated by SME’s (small & medium sized enterprises) are not connected at all, even large industrial operations use conventional equipment. And simply connecting to legacy equipment is a very difficult and investment-heavy process. The cost vs effectiveness curve of converting legacy machinery into intelligent assets is a deterrence indeed.

Data Format & PLC Issues on Smart Equipment

Ignoring the legacy equipment pool, reaping benefits of advanced data analytics on smart industrial assets is no easy task. Although smart assets are designed to capture data, these data are only relevant for the asset producers’ (OEM’s) equipment design architecture and logic.

Although smart assets are designed to capture data, these data are only relevant for the asset producers’ (OEM’s) equipment design architecture and logic.

The data’s format, frequency, content, or size do not necessarily satisfy requirements for advanced analytics that would generate valuable insights for operational processes. To make things even more difficult, same assets that are designed and produced by the very same OEM may have different configurations. Therefore, they may speak different PLC languages. Those need to be translated to a common framework by the owner-operator. Assuming PLC data are converted into a common analysable framework, there are still other pitfalls to be considered. For instance, separate equipment times may not precisely match. It may sound like there should be an easy adjustment, however it is possible only if the problem gets noticed. Working with two assets that work on the same production line, algorithms may obliviously compare mismatched time series data (real life, unnoticed issue).

Cloud vs On-Prem

Data need to be managed either on servers (such as on-prem OPC servers) or on cloud platforms. Cloud solutions are more AI-compatible as petabytes of data can be managed effectively. However, cloud computing is admittedly not for everyone. Most industrial operations still refuse to take data outside borders. Some categorically reject cloud (yes, it happens), while others cannot even consider it due to regulations. If on-prem is the only plausible option, it goes without saying that all cabling and data storage infrastructures would have to be engineered and commissioned in the first place. That unfortunately comes with other inherent issues and exorbitant costs. On-prem servers can hardly manage enough data effectively to realise the potential benefits offered by advanced analytics.

On-prem servers can hardly manage enough data effectively to realise the potential benefits offered by advanced analytics.

People & Skills

We argue it is all about data integrity, which is correct. But finally, people and skills need to be considered besides infrastructural aspects. Employing highly skilled data scientists to integrate data, hardware and software that produce meaningful outcome is a significant challenge. Most industrial companies either lack the human resources or the patience, most likely both. On the other hand, data scientists, who have production industries background, are rare. And there is often a large mental gap between AI experts and traditional operations people. There is even a larger gap between AI experts and incumbent Information Systems silos, with whom the new generation resources are expected to collaborate.

There is even a larger gap between AI experts and incumbent Information Systems silos, with whom the new generation resources are expected to collaborate.

As a summary, data science and AI offer a great leap forward for production industries. However, those industries lack the infrastructure to ensure data integrity as a pre-requisite. Besides, human resources with the right skills and experiences are rare. Therefore, we believe the industry's AI journey needs to be approached with caution. Until the fundamentals are addressed, the foot taking the leap will remain in the digital airspace, while the other remains on soft ground.