Predictive Maintenance in Cooling with Machine Learning techniques

Organization

Sabcobel is a refrigeration company which provides cooling elements for both the retail and  industrial sector. The company was founded in 1958 and today has around 80 employees  spread across Belgium and Luxembourg. They offer an end-to-end cooling solution to their  clients, from design to implementation followed by a 24/7 monitoring of their systems. 

Challenge

Sabcobel has hundreds of cooling installations to monitor and control. Each installation is critical as it controls the quality and lifespan of the  products it contains (e.g. food  products); as a result, the cooling  needs to be closely monitored and -  when temperature surges occur -  timely addressed. Every cooling installation has sensors  attached to their cooling elements in  order to trigger an alarm in case of a  defective cooling element. This  allows Sabcobel to make a quick  intervention to fix the defect.  However, every defect results in  repair activities as well as downtime  costs for the customer which are costs they want to minimize. Therefore, Sabcobel was looking to change  their  maintenance organization  from  this  break-fix model  to  a  prevent-optimize model where they would predict the failure of a system and perform a preventive fix in order to minimize the repair costs and downtime for their customers. 

Solution

Sabcobel worked with element61 to  set up a system to have

  1. a real-time monitoring system for all their assets and
  2. predictive maintenance in place to predict installation failures & alerts before they occur.

Initially, a monitoring system was put  in place using Microsoft PowerBI to  give the customer interactive  reporting capabilities with  proactive notifications of failing  systems. Secondly, using all historical  data available, a Machine Learning  prediction model was built in  Microsoft Azure to predict when an alarm would be triggered, but also what the cause of an alarm would be. The latter was an important factor so Sabcobel is able to understand whether a predicted alarm would  really result in a failure and a  technician needs to be sent or whether it could be easily solved remotely.

Predictive Maintenance in Cooling with Machine Learning techniques

Figure 1. example of an interactive dashboard

As part of the predictive tool, element61 has built in a feedback-loop which is used to indicate whether the predictions were correct or not. This important loop is incorporated in the solution to leverage the customer input towards  improvements in the model performance. In order to guarantee that the tool is running continuously, we put a monitoring mechanism in place informing us in case of a failure of the prediction tool.

Result

By considering their data as a real valuable business asset, Sabcobel has been able to transform their service organization to a prevent-optimize organization. Sabcobel is now able to reduce their maintenance costs as well as limited the downtime for their customers which increases customer satisfaction.

“During the hot summer 2018,  the monitoring system built by element61 already proved more accurate & responsive than our previously used  alarm call center. This means the quality & responsiveness of our service is guaranteed even when it’s busy & pressure is high.”

- Matthias Coppens, Service Manager Sabcobel