Audience Opinion Piece

Artificial Intelligence (AI) is transforming industries at a rapid pace, and the manufacturing and industrial automation sectors are no exception. As companies increasingly turn to AI to optimize processes, enhance productivity, and predict failures, understanding the data behind these solutions becomes crucial. Three key factors—data acquisition, data aggregation, and data ownership—are central to unlocking the full potential of AI in this space. Let's delve deeper into these concepts.

‍

Data Acquisition: The Foundation of AI-Driven Insights

The first step in building AI solutions for industrial automation is understanding how data is acquired and, more importantly, the context in which it is collected. Data acquisition isn't just about gathering large amounts of information; it’s about capturing the right data that can provide meaningful insights. This is where system integrators and solution providers come into play, including the original equipment manufacturers (OEMs) of machinery. Their expertise is crucial in determining which data points are critical to monitor and what those data points signify.

‍

For example, sensors may track various metrics such as air filter status, oil temperature, or bearing wear. Each of these data points tells a story. Is an air filter approaching the end of its life cycle? Is oil temperature creeping beyond safe thresholds, signaling potential overheating? Are bearings showing signs of wear that could lead to equipment failure? The sensors or programmable logic controllers (PLCs) capturing this data can provide early warnings of potential issues, allowing manufacturers to take preventative measures before they escalate into costly downtimes.

‍

However, acquiring this data requires close collaboration with the OEMs or other solution providers who have an in-depth understanding of the machinery and its operational parameters. Without this collaboration, the data acquired may lack the context needed for AI algorithms to generate actionable insights.

‍

Data Aggregation: Bringing It All Together

Once the critical data is acquired, the next challenge lies in aggregation. In an industrial environment, data comes from a multitude of sources—robot controllers, PLCs, databases, sensors, torque controllers, and more. Each of these sources may use different communication protocols, formats, and standards. Aggregating all this data into a cohesive and unified stream is essential for ensuring that AI and analytics solutions work effectively.

‍

A significant challenge in data aggregation is normalizing the data. Different data sources may have different timestamps, units of measurement, or even definitions of what constitutes a “failure” or an “anomaly.” Without proper normalization and conversion, AI systems may lose context, leading to inaccurate predictions or recommendations. For example, if torque data is measured in Newton meters in one system and foot-pounds in another, failing to normalize this could lead to erroneous conclusions.

‍

Moreover, data flow is also crucial. Understanding where data will flow—whether to cloud-based systems, edge devices, or on-premises servers—affects latency, storage requirements, and, ultimately, the speed at which AI can process information and deliver insights. Seamless integration across data sources ensures that AI models have access to all the relevant information and can perform at their best.

‍

Data Ownership: Who Owns Your Data?

One of the most critical but often overlooked aspects of AI in industrial automation is data ownership. The industrial automation space is no stranger to vendor lock-in solutions, where proprietary industrial protocols, unique connectors, and closed systems make it challenging to switch vendors or integrate third-party solutions. Unfortunately, the same risks exist with AI solution providers.

‍

As AI becomes more accessible and the hardware needed for these systems becomes more standard and affordable, companies are increasingly looking to build AI solutions in-house. However, many AI solution providers offer models that provide valuable insights and outcomes without allowing customers to access the raw data used to train these models. This can create a new form of lock-in, where companies are dependent on the provider’s AI models rather than owning and controlling their data.

‍

The implications of this can be significant. Take, for instance, a non-climate-controlled manufacturing facility that produces automotive interiors. Seasonal temperature changes could lead to increased failure rates for adhesives used in these products. An AI model trained on years of production data might eventually recognize this trend and recommend changes in the process. However, if a company is locked into a vendor's AI model without access to the underlying data, they may struggle to replicate this insight if they ever switch providers or attempt to develop their own AI models.

‍

Having a clear data strategy is essential. Companies need to ask themselves critical questions: Do we own our data? Can we export it and use it in other systems if necessary? Do we understand the trends and insights that our AI systems are generating? Data ownership isn't just about retaining access to raw data—it's about controlling the insights that drive operational decisions.

‍

Conclusion

AI in industrial automation offers immense potential, but its success depends on a robust foundation built on understanding data acquisition, aggregation, and ownership. By collaborating closely with system integrators and OEMs, ensuring seamless data aggregation, and maintaining control over their data, companies can harness the full power of AI to drive efficiency, productivity, and innovation in their operations.