Lead times are a hot topic these days. It’s not at all uncommon to have to wait 4 or more months for components or spare parts that until recently could be delivered within a single week. Spare parts are essential when maintaining a site operational. Today however it thus isn’t uncommon for a production line to be out for a few weeks because critical components are missing.
A classical approach to mitigate such a situation is making sure to have all components that can fail in a strategic stock. As such; no waiting times, as the parts are readily available. In practice, this is however often an unrealistic approach due to the associated capital cost and operational impact. In an ideal world, one would receive a warning weeks before a component fails such that replacements can be prepared, or failures can even be prevented by fixing the cause of the observed degradation.
Failures resulting from external impact or human errors are of course hard to predict. A significant proportion of issues however results from slower processes or gradual changes. It is often not the component degrading, but the context the component operates in changes or evolves, resulting in higher stresses or altered behavior. These are the situations where context-aware monitoring brings a lot of added value. This specific type of follow-up is based on continuously collecting data from multiple sensors of different types on a multitude of locations on the installation or asset. In practice the different sensor types that are used measure electrical currents and voltages, while others measure for example vibrations in multiple axes.
Based on knowledge of the asset a model is established that continuously evaluates the ensemble of data coming in and checks if the behavior of the components is normal or not. The model has to be rather specific for the asset as it:
- Is built up of pre-existing knowledge of physical relations associated to the type of asset.
- Also contains an AI-based aspect, trained on historical data on the same or similar assets.
- Derives and takes into account the status or operating condition of the machine: what is it doing (e.g., starting up in regime, making product A).
- Offers context to the end user: alarms and warnings are associated with an explanation or background on the origin of the deviation observed or process taking place.
Such an approach may sound like a complex endeavor. Luckily pre-made models exist for multiple installations. They can be applied to a new asset of the same type, will bring value from day one, and will grow more intelligent and specific for your machine when time progresses.
These kinds of approaches result in a follow-up where upcoming damage or unwanted operations are detected in an early stage. The owner receives a warning well ahead of time and, based on the context added, can initiate the appropriate action: repair or replacement. In the end: the most sustainable component often is the component you don’t have to repair or the machine that has the highest availability.