Step by step to predictive maintenance with minimal project risk

The idea behind predictive maintenance sounds promising: The sensor data of a plant are used to determine the time and type of necessary maintenance work. In this way, maintenance processes can be optimized and costly, unplanned downtime reduced. However, the structure of a predictive maintenance solution is complex, associated with new technologies and thus represents a project risk that should not be neglected. A step-by-step approach, starting from the Minimum Viable Product (MVP), helps to minimize the risk. This will be explained briefly using an example.

Initial situation

Authors: Floarea Serban & Geri Reif

New technologies often fail because the approach to their implementation is not optimal. The complexity is enormous and the imponderables, coupled with a lack of experience, are often daunting. The appropriate approach, namely a step-by-step project implementation, can significantly reduce a large part of the risks. Because by having an executable product at each step, it can be decided from case to case what the further procedure will look like.

Infrastructure and visualization using an example from the construction industry

To illustrate this, we look at a predictive maintenance case in the construction industry. Here, construction machines are equipped with sensors to send data such as oil level, GPS coordinates, battery voltage, operating hours, etc., to an IoT platform. In the IoT platform, in the first stage of expansion, the data is forwarded to a database for storage and clearly displayed in a dashboard. In this stage of expansion, the building yard has an overview of the location and operating status of the construction machinery at the push of a button.

Fig. 1: Overview of the location and operating status.

Machine learning and pattern recognition

In the final step towards a predictive maintenance solution, the sensor data in the database is examined for anomalies that have led to problems in the past using pattern recognition and machine learning. If such patterns are detected, the Stream Analytics component is configured to detect these problems in the incoming sensor data in real time and to trigger an appropriate reaction. In this way, it is possible to react to critical operating conditions before major damage occurs.


In this blog post an example was used to show how to implement a predictive maintenance solution step by step. Each additional intermediate step generates additional business value. The MVP will be successively expanded and the experience gained with the new IoT technologies will significantly reduce the project risk. Maintenance work will become more efficient and unnecessary costs will be eliminated.