Case Studies Machine Learning Proof-of-Concept Project for Manufacturer

Machine Learning Proof-of-Concept Project for Manufacturer

Our client engaged Concurrency for their first data science project, focusing on leveraging machine learning to lower costs and enhance operating efficiencies. Through a series of meetings and data exploration, we identified machine-tool and machine-operator efficiency as areas in greatest need of improvement, leading to the development of a hypothesis and methodology for the proof-of-concept outcome.

Our client requested Concurrency’s assistance with its first-ever data science project. In this proof-of-concept project, we examined where and how machine learning could help lower costs and improve operating efficiencies. 

In machine learning proof-of-concept projects, we hold a series of meetings with our client’s business leaders, process managers and product managers. The questions we ask help to identify which business processes are both in need of improvement and a fit for machine-learning solutions.

In this case, our client identified general machine-tool and machine-operator efficiency as the greatest need. Our client incurs costs when machines suffer downtime, which can result when a tool fails. Tool failure also leads to waste of expensive raw materials, as in-process machined parts must be scrapped.

In our workshop sessions, we explored data our client has available to describe processes relating to machine-tool downtime. In general, machine learning may be a good fit for scenarios in which substantial historical data is available. Having identified that data, we then carried out data exploration. We assembled a series of reports to describe the data’s distribution. By analyzing its variance and other statistical measures, we explored ways the data might violate an intuitive understanding of the machine-tool manufacturing process.

We followed this analysis with additional discussion with our client’s business and technical leaders to gain more insight into the manufacturing processes. From that point, we’re in a position to move into hypothesis generation and, in turn, methodology generation to guide work beyond the initial hypothesis and toward an implementable model.

In this particular case, our hypothesis development led us to focus on the timing of machine-tool replacements—that is, at what point does an operator replace a worn tool. This hypothesis and related methodology (to guide further work) represents the proof-of-concept outcome for an initial machine learning project. Next steps beyond proof of concept include detailed modeling, testing and implementation.