A Decision Needs to be Made
I have written here before about the some of the pitfalls with machine learning and artificial intelligence. There continues to be a misunderstanding about capabilities of artificial intelligence and allowing it to make difficult decision. As noted before, it is critical to understand how well your model fits the real world conditions. Moreover, the result needs to make sense. Artificial intelligence will provide some great insights, but humans will still be needed to ultimately make the decision.
Weather it will Rain or Not
Weather has long been a topic of office conversation. The local weather person is often the source of much frustration and the target of criticism. Who else can be wrong more often that right and continue to succeed in a profession? The challenge of course is the ability to forecast. The models are certainly much improved, but with all of the factors that go into weather, it is quite difficult. The meteorologist will interpret the data and provide a forecast based on that analysis. That decision leads to a decision for you.
Because forecasting (and artificial intelligence) relies on probabilities, results can exist anywhere from no chance to a one-hundred percent chance. Where this impacts you is how you prepare for the forecasted weather for the day. For example, should you take your umbrella to work. It is big and bulky and can be a pain to deal with. If the forecast calls for a 90% chance of rain, you will likely take it. One the other end, if the forecast calls for a 10% chance, you will likely leave it at home (when it rains, of course, you can blame the weather person). But what if there is a 40% chance? Or a 60% chance? This is ultimately the decision you will have to make.
Good and Bad Vibrations
Machine leaning and artificial intelligence are commonly used in process reliability. As the P-F curves tells us, once a potential failure is detected, there is a period of time to a functional failure. The goal of a reliability program is to not only detect the problem sooner, but better predict the exact time of the functional failure. These are all based on probabilities which use previous events to predict future ones. Like investing, previous performance is not a guarantee of future results.
Vibration is a common root cause of rotating equipment failure. A small vibration usually indicates there is ample time to failure. Significant vibration means the time to failure is in the near future. You must now decide when to make the equipment repair. This can be tricky due to maintenance demands. The repair may require a process shutdown, and these can be costly.
Whether to Shutdown or Not
Decision making will be required when the predicted time to failure falls within a planned shutdown event. Similar to the umbrella decision, if the predicted time has high confidence, you will likely wait until shutdown. If is has low confidence, the repair will be performed before the shutdown (you may want to consider making other repairs at this time to mitigate the time needed for the planned shutdown). But what is there is only a 50% change the equipment will not fail before the shutdown?
Should you find yourself in the last position, I encourage you to have a conversation with all stakeholders (a good reliability program, like Reliability Centered Maintenance, will enable you to do so). This will allow you to have the conversation so that everyone can discus the situation and agree what to do. As I mentioned, this may lead to moving up the shutdown. Or you may be decide to take the risk (a proper risk assessment will aid in this decision). Either way, as the reliability person, you don’t want to become the weather person.