There are so many predictive maintenance solutions out there. How can you know which one is best suited to your needs? When it comes to choosing the right solution for predicting failures and preventing costly shutdowns, it is important to distinguish between process and discrete manufacturing. For some of us it is common knowledge, yet for others the differences are not so obvious.
Whereas both types of industries aspire to seamless production, preventing shutdowns, lowering maintenance costs and improving the efficiency altogether, we suggest that you consider the following points for the process industry:
1. The data exists
The process industry is heavily regulated. When we hear of regulation, we usually think of a barrier to innovation, something to stop it from penetrating the industry. But in the case of predictive maintenance software, it actually facilitates the access. The manufacturers were required to invest huge amounts of money to implement sensors and collect data, for regulatory and operational reasons. It is very difficult to control the plant without these measurements. Now that the data exists, it makes a lot of sense to look for solutions that will leverage the data to the benefit of the plant.
2. It is one continuous and complex process
Since the procedure in discrete manufacturing is divided into several fragments, it is easier to locate the problem, determine what went wrong, stop production and fix the problem. In the process industry, the production is more complicated, and a strange sensor reading in one place could indicate a failure in another place that seemingly has no connection. For this you need a sophisticated technology, one that “sees” the entire plant and is able to make the connections between the different parts and different processes.
3. There are thousands of sensors
Another important difference is the number of sensors that the process industry is dealing with. The plants of the process industries have thousands, sometimes tens of thousands, of sensors. Having so many sensors means that there are many possible normal states. It also means that at any given point in time there will be an abnormality in the data. The software has to identify when the system deviates from the normal state, and most solutions manage to do that with machine learning. But the abnormality in the plant is not per sensor; whether a situation is normal or not, is a derivative of an aggregation of sensors. Possibly, per sensor the situation could be abnormal, but the aggregation of all the sensors together could point to a normal state. A good predictive maintenance solution has to bridge this gap between abnormalities in the data and failures in the plant.
4. Few historical events exist
Not only is the number of sensors an issue when choosing predictive maintenance software, but also the number of historical failure events. Unlike the discrete industry, in the process industry there aren’t enough historical failure events to provide a good prediction model for each type of failure. For example, if we don’t have enough occurrences of a crack in a boiler, we can't assume that the pressure reduction we have seen in all of the historical cases is indeed connected to the crack, and not just a chance co-occurrence of random values. In order to identify with reasonable confidence that this is indeed an indication of a problem caused by a crack, we must have enough examples of past occurrences of the problem. The lack of a large enough number of historical events makes solutions based on prediction models unsuitable for the process industry.
When considering a predictive maintenance solution for the process industry, you have to think about the unique characteristics of this industry. You have to take into account that the data already exist, that you are dealing with thousands of sensors, that the procedure of manufacturing is more complex than for other industries, and that there are few historical events. Thus, the solution required has to be more sophisticated, make all the connections between parts and processes, and be able to bridge the gap between anomalies in the data and failures in the plant.
We at Precognize cater precisely to these criteria. Precognize adds a layer of knowledge of the plant on top of an excellent machine learning algorithm. In an easy-to-use studio, the operation people describe the structure and the processes to create a model of the plant. Adding cognitive intelligence to the data results in a few pinpointed alerts, ahead of time, thus predicting failures and preventing costly shutdowns.
Learn more about our solution: REQUEST A DEMO