Putting Predictive Maintenance into Practice
How effective is predictive maintenance? How can IIoT solutions predict machine downtime while also taking timely action? These weighty questions are answered via this ei3 research paper that analyses the use case of a premium printing machine builder. The article also details the approach and initial results while identifying improvement areas.
ei3 REPORT ON FACTORY FLOOR EXPERIENCES IN FORECASTING MACHINE DOWNTIME BASED ON REAL-TIME PREDICTIONS OF IMMINENT FAILURES:
- The paper begins with an in-depth look at the machine that serves as a test subject – the Central Imprint Flexo Printing Machine. The device is mainly used for small-batch printing jobs, resulting in high downtime rates due to component changes – which is ideal for the purposes of looking at how predictive maintenance can help reduce downtime. The machine is equipped with several sensors to monitor crucial data for analysis.
- The next section analyses the down-time and process data, with several error codes belonging to each data segment. The process data consists of as many as 93 variables. Along with machine settings, these were collected and correlated with the machine slowing down in cycles of approx. 17 minutes. It showcases just how impactful operator behavior can be while monitoring device data and other environmental factors.
- After data collection, we dissect the process with which a prediction algorithm for this machine might be developed. Prediction horizons, event choices, delimitations and more are considered, all of which feed into the creation of this algorithm.
- The algorithm is developed. It is configured to predict print unit failures instead of other, less practical options like focusing on operator errors that are largely to do with non-objectifiable factors like human behavior. A suitable option was developed after a multi-stage process that involved feature engineering and algorithm selection, while also considering a variety of false positives and negatives.
- This section concludes the study where the merit of predicted failures is seen despite a host of other unforeseen challenges witnessed in this particular study. The takeaway? Taking some measures that allow for predictive measures might be challenging, yes – but it is ultimately effective, and might even be necessary as the industrial ecosystem evolves.