Not every production plant is the same. Within a site, not every installation is the same. When working in manufacturing the goal is to keep the availability of each machine on a maximal level. For critical machines, these used continuously or with spare parts that are hard to come by, this is specifically important. A maximal uptime can be best obtained based on a continuous, data-based monitoring of the asset and its behaviour. The approach that is the most appropriate for an individual installation differs however by… the simplicity of the situation. Here’s how:
- Simple machines in this context are those that are running continuously and at (almost) the same speed. In this case one can track the behavior using electrical or vibration sensors and a mere anomaly detection on the data generated can suffice to detect that something is going on. Such a detection will trigger a signal and a maintenance engineer can go and check the issue. If required a repair is planned in. For this type of situation a simple, independent IoT sensor delivering short blocks of data on an hourly basis is a perfect solution for a good follow-up and to keep the line up-and-running.
- More complex situations are met when the production line or installation consists of multiple different components that are not necessarily working at a constant speed or continuously. The source of these variations can be in an intermittent or irregular operation, different regimes, or multiple materials that are produced or processed in the same installation. If something happens the situation is not always “simple” either. Therefore it is required that the monitoring solution is aware of these changes and also provides the necessary context to the maintenance engineers. This:
- Helps to avoid false positives
- Is essential to be able to set sufficiently sensitive boundary values
- Allows the mainenance team to start the job in a targeted way
- And gives them all the tools to reduce the duration of the standstill to a minimum (or even avoid an intervention during production hours)
In the case of a more complex situation, a proper monitoring solution should consist of different sensor types generating a (quasi) continuous data stream, combined with an equaly continuous stream of process-related data ingested from controllers or historians. An approach based on mere anomaly detection won’t suffice at all either. The incoming data needs to be classified, correlated and transformed into descriptive parameters and more complex models are required for early detection as well as contextualization.
In reality, in most manufacturing or production sites both situations are met. This means that in a realistic case a hybrid approach is followed, depending on the properties of the asset and the properties of the operations. Up to you to decide where to start first…