I recently put together a predictive maintenance machine learning (ML) solution for a client. The goal is to optimize the allocation of a limited maintenance budget so that system down time is minimized.
The client had 10 years of data on system failures and maintenance. I built features based on this data and additional ‘unit’ attributes. Using these features and the historical failure information, I trained a binary classification model to predict the likelihood of a unit failure in 2022. Finally, I built a Power BI report that visualized the probability of 2022 failure for all units along with the input features for context / explainability. The planner will use this report as a primary input to the 2022 maintenance planning.
This was a pretty basic ML application. The tabular data wasn’t big and it was relatively clean. While I tried a few different algorithms, the winning algorithm was quite basic and interpretable. There were no data privacy concerns. There is a human in the loop that is ultimate decision maker. And it’s not a scenario where predictions need to be continually refreshed nor does the model need to be regularly re-trained.
ML can get complex, but it doesn’t have to be. And it doesn’t have to be expensive; this solution certainly wasn’t for my client. But don’t take my word for it. An AI leader, Andrew Ng, wrote an article for Harvard Business Review last year with exactly this theme – AI Doesn’t Have to Be Too Complicated or Expensive for Your Business (hbr.org) (https://hbr.org/2021/07/ai-doesnt-have-to-be-too-complicated-or-expensive-for-your-business).
There are several paths to a getting a great return on your analytics investment. Message me if you’re interested in starting a discussion about your challenges with me.