By: Humera Malik
I recently participated in a panel discussion at OSIsoft’s PI World on the promise and pitfalls of Machine Learning and Artificial Intelligence. I thought it would be great to share some of the top questions asked during the panel to determine how to know if your organization is AI ready. You can also view a video of the panel discussion here.
A lot of time is spent debating the definition and difference between Artificial Intelligence and Machine Learning. At a simplistic level, AI helps machines to perform tasks in intelligent ways by having the ability to adapt to different situations and new data. Machine Learning is seen as a subset of AI and enables machines to process data and learn on their own, without constant supervision. However, it’s important to understand in the industrial plant context – AI and Machine Learning are tools to add to your operations toolkit. In order to create powerful models, however, they need contextual data and intelligence that only operations experts can provide.
What is often underestimated is the time spent on preparing the data for AI applications. It’s been suggested that two thirds of an AI project is spent on data preparation to really make AI fly and derive the valuable insights. In the best-case scenario, time series data in a structured manner is required to expedite your AI project. We need enough data to train our models as well as data that will help the model to recognize failure.
Advanced Process Control (APC) models are built using years of experience (heuristics), or in control experiments (empirical) to come up with formulas that are specific to each operation. In most instances, APCs draw on one or two main data sources to understand and address particular performance improvement opportunities in a process or asset. However, where there are external variables or datapoints that impact the process or asset, such as environmental changes, this is where the value of AI comes into effect. AI is better at coping with changes in the circumstance and can provide a 360-degree view of the process. In contrast, while APCs are fed multiple datapoints, they still make decisions based on the specific model. The limitations with APCs mean that the model may be making decisions without the complete story and it doesn’t have the ability to ‘re-learn’ with new data. The opportunity with AI is to combine heuristic and machine learning models to create very powerful models that utilize historic and new data to optimize decision making.
"At Canvass we work closely to build confidence in what AI can deliver to an industrial operations plant."
While you may see the opportunities for AI are vast, getting your boss on board may seem like an uphill battle. That’s why at Canvass we work closely to build confidence in what AI can deliver to an industrial operations plant. First up, we map out the operational outcome and how success will be measured. By having the metrics in place, we will be able to avoid any ambiguity between what the model has produced vs the status quo. We then take the time to understand the impact each attribute has on the process and use our model to validate the assumptions or derive new insights that will impact the outcome. It’s a collaborative effort between our team and your subject matter experts, as no one knows your business better than you do. Where AI comes in is to uncover opportunities for improvement that are difficult for the ‘naked eye’ to easily identify but will be impossible for your boss to dispute the value.
