- A medical device company has maintenance contracts for thousands of their instruments installed in hospitals and health facilities. They wanted to reduce the number of technician repair visits by looking for trends in the data these machines collect. But they were uncertain if the data collected by these machines was sufficient to support predictive analytics. Furthermore, most of these machines were not connected through the internet, so a justification for an investment to retrofit the machines was needed.
- In a proof of concept, with holdout analysis, we built prototypical machine learning models to relate error codes with subsequent system faults. The ability of the models to predict the needs for costly technician service calls was a factor in making the business case justification for investment.
- It was ultimately determined that it was too costly to retrofit older machines, but that future machines should be designed with data systems supporting predictive analytics. This is a common outcome of IoT analytics proofs-of-concept.
- However, a result of the project was that we determined, through service record history, that scheduled maintenance calls could proactively replace components likely to fail soon. This results in reduced labor and parts costs, and higher uptime for the customer.