A Multinational Mechanical Manufacturing business, present in over 20 countries and with more than 10 production plants. The business has approximately 4000 employees.

The overall objective of the business was to become data-driven. The first challenge to overcome was to optimize the production lines and to detect abnormal behavior of the machines producing the mechanical components in real time to anticipate possible downtime on a production line.

This was particularly challenging due to the high variety of machines and components made during the production processes. This variety makes it hard to find a ‘standard’ approach to detect the data patterns across all machines and production lines.

Identify the best Machine Learning approach able to detect anomalies that can predict machines downtime. This approach is able to provide an anomaly score based on historical data collected (exploiting the collection system already put in place by the customer). Based on the anomaly score it is possible to identify a threshold above which the behavior can be classified as abnormal.

Using a Machine Learning model, the output signals can be compared with the input signals. If the difference of this data exceeds a certain threshold, anomalies are detected.

Each anomaly is displayed on a dashboard showing an overview of the plant which can be immediately seen by the relevant workers.  This instant visibility and advanced knowledge empowers the teams to plan ahead and anticipate downtimes to make appropriate amends.

Our client has seen significant advancement in the monitoring of the production lines. Thanks to this project our customer also achieved the following:

 

  • Reduction in downtime caused by sudden stops
  • Reduction in maintenance costs
  • Increased understanding of machine use
  • Optimized use of the production tools lifetime
  • Optimization of spare parts