Big Data and Machine Learning is very good at finding correlations between large numbers of sensors and equipment breakdowns or quality issues. This is incredibly valuable if you have a very complex machine where the reasons for a failure are difficult to identify.
The Big Data and Machine Learning approach to predictive maintenance looks like this:
- Install lots of sensors to learn as much as possible about your equipment
- Collect massive amounts of data for a long period of time
- Run computer algorithms that correlate the sensor data to machine failures
- Use these results to predict future failures
The up-front costs of this approach is high, the return on investment is slow, and few companies are willing to suffer through the time and expense of numerous failures in order for the machine learning algorithm to learn the leading indicators. A faster and more cost effective approach is to use a simulation model for your equipment. Chances are you are not the first company trying to avoid downtime on a pumping system, assembly line, extruder, etc. , so there is no need to wait for a machine learning algorithm to derive the correlations between the sensor data and the eventually failures. Simulation models combine modelling algorithms with specific application knowledge in order to predict failures or quality defects. This approach requires fewer sensors and can be fully implemented in weeks rather than months or years. The predicted failures are calculated from day 1 with increasing accuracy over time. The lower cost and fast results provide an extremely compelling return on investment for industrial equipment manufacturers as well as the end users. This session will cover: approaches to predictive maintenance (machine learning, simulation models, and rule based models); advantages, disadvantages, and ROI for each approach; case studies; and recommendations for getting started.