How AI Determines Correlations in Data

In our new series, "Can AI do that?", we address common questions and misconceptions about AI.

At VROC, we prioritize holistic data and request two years' worth of historic operational data for model training from various systems, including DCS's, Historians, SCADA systems, direct sensors, and third parties. However, a common question often arises:

"Can VROC AI distinguish good data from bad data?"

To delve deeper into this issue, we spoke with Trevor Bloch, VROC founder, who explained that AI doesn't focus on the quality of data but on the correlations between specific sensors and the process or equipment being optimized. This means that even if a sensor is producing inaccurate values or isn't calibrated, AI can still identify a correlation as long as the values change in conjunction with the process. For example, if a temperature sensor is producing values that change in accordance with the temperature changes, AI will recognize the correlation and use the data.

However, if there is no correlation between the data, AI will disregard it. In this way, AI can filter out bad data without necessarily knowing whether the sensor is faulty or not. By focusing on correlations, AI can identify patterns and relationships that human analysts may overlook, leading to more efficient and effective optimization.

In industrial continuous processes, where large amounts of data are generated from various sources, it can be challenging to discern good data from bad data. But by understanding that AI's primary focus is on correlations, we can move beyond the misconception of good versus bad data and focus on the correlations that matter, and leverage the power of AI to optimize our processes.