Predictive Maintenance Use Case

Challenge

Industrial organizations require equipment to run continuously. Before the adoption of digital manufacturing practices, factory managers were required to manually schedule (corrective or preventive) maintenance operations and repair machine parts. Corrective maintenance where parts are replaced when they fail costs downtime and unscheduled maintenance. On the other hand preventive maintenance might avoid catastrophic failures but it also leads to under utilization of the replaced components and unnecessary downtime. Predictive maintenance is a proactive maintenance method looking to predict machine failure and proposes to balance the two approaches by enabling just in time component replacement. According to McKinsey & Company, predictive maintenance “promises a very tangible benefit: machines that don’t break down”.

Predictive Maintenance is helping factories achieve Smart Factory efficiency and status and it is being chosen as a go-to high-value digital asset management solution by many organizations, as it has the potential to drastically reduce downtime (our solution helped decrease downtime by more than 25%, see below), to reduce maintenance planning time by 20-50%, maintenance costs by 5-10%, implement operational cost savings by 5-10%.

Our Solution

For one of our customers we developed a thermal imaging-based Predictive Maintenance Solution for operating electrical equipment. Thermal imaging allows for measuring temperature of imaged objects by collecting invisible infrared radiation (see a sample in Fig.1).

Thermal data was automatically acquired alongside geolocation data and further sent to our system. Then our solution was able to analyze thermal images, detect thermal anomalies and raise an alarm for our client experts to validate the findings. We delivered a dashboard allowing for visualizing data (i.e. historical data or location requiring expert inspections)

Fig. 1 - Industrial Thermal Image Author: Dylan Pankow, Source: Flickr, Changes: No

Solution Details

Our solution can be decomposed in 2 parts:

• A data acquisition and transformation system whose responsibility is to automate data acquisition and data processing. Once data enters this system, it is automatically registered and stored. The thermal imaging data is then preprocessed along with some metadata (i.e. location or datetime) in a typical ETL process. In order to measure the electrical equipment temperature, a deep learning model is trained to detect the location of this equipment. Once isolated, the distribution of the equipment temperature is extracted, and can be compared with previous measurements to detect the apparition of hotspots. Hotspots are a sign of component degradation. In addition, a more subtle anomaly detection is performed in order to report abnormal components.

• A reporting and visualization tool (dashboard) allowing the end user to be informed about the state of the electrical assets and engage further initiative like part replacements.

In addition, this dashboard introduces a human-in-the-loop: the end user can contribute to the efficiency of the data analysis system by correcting wrong inferences.