Artificial intelligence for predictive maintenance at our facilities
Historically, organizations deployed a reactive maintenance strategy, carrying out repairs when a fault arose. Subsequently, other maintenance techniques emerged.
At our facilities, maintenance is one of the most demanding tasks in terms of both resources and time.
The commonly accepted maintenance strategies now are:
Corrective maintenance: maintenance related to less important tasks or lower-cost repairs. Under this strategy, a machine can work until the point where it fails.
Preventative maintenance: this maintenance regime is used in different situations, the most common being with equipment that has long response cycles and no standby mode. Interventions are undertaken even when the equipment is working satisfactorily.
Predictive Management (PdM): predictive maintenance is a maintenance strategy driven by predictive analysis. Technology is used to detect patterns of failure and anomalies and to estimate the length of the interval before a malfunction is likely to occur.
Prescriptive maintenance: this is based on a combination of maintenance management tools and predictive strategy. This allows us to take decisions based, not only on the condition of the equipment, but also taking into account economic factors, production, etc.
The emergence of technologies like Big Data and artificial intelligence allows an innovative focus for predictive models.
At our facilities, we store process data in our Data Lake so that it can be monitored, and algorithms applied based on the performance of the task.
In addition, some new equipment is fitted with sensors so that its condition can be monitored in real time. In many cases, monitoring by the sensors is sufficient to detect anomalies before a malfunction occurs.
Maintenance has become a key task at our facilities. Improvements in the technology used in sensors (which has made them cheaper, easier to fit and more reliable) will provide us with even more data in the near future. This will enable us to apply new algorithms which will help us to reduce even further unplanned downtime in our equipment and installations, thereby increasing uptime at our facilities and reducing maintenance costs.
It should be borne in mind that having this data available, and being able to view it, in itself makes a great contribution to identifying inefficient operations or performance and to producing new algorithms which will predict faults automatically.
In order to achieve this new focus on predictive system maintenance, our data scientists and engineers (those based in workshops, and those concerned with reliability and inspection, etc.) will need to work together on data analysis and on the generation of models.