- Manufacturers should move beyond hours-based maintenance scheduling to cycle-based tracking using MES production monitoring, which provides more accurate maintenance timing.
- The optimal approach combines preventive maintenance (using planned cycle counts to prevent problems before they occur) with predictive maintenance (real-time process monitoring of temperature, pressure and voltage) to create an effective safety net.
- MES historian databases enable four critical predictive maintenance processes — high-low boundary alerts, trend analysis, pattern matching and sum of evidence techniques — that can incorporate AI tools to detect potential failures before they impact production.
I recently overheard two plastics manufacturers discussing their operations. One said, “I noticed you schedule tool maintenance based on hours of operation rather than actual production cycles.” The second manufacturer responded, “You’re right; we should track cycles.”
That plastics manufacturer is far from alone. Despite understanding the advantage of monitoring production cycles, many companies are deterred by the perceived complexity of collecting that data. Instead, they still base their maintenance scheduling on hours of operation.
However, managing tool and equipment maintenance with hours is a bit like scheduling the maintenance of a vehicle based on mileage. It is a great way to time oil changes, but on modern vehicles, the crucial maintenance indicators are the sensors and analytics built into the vehicle.
Similarly, most modern shop floors have a wealth of data available from smart machines, sensors and even simple cycle counters that can be used to predict when maintenance will be required.
The key to unlocking this information lies with the production and process monitoring of a manufacturing execution system (MES).
Combining preventative and predictive maintenance
MES-based production monitoring counts usage cycles on tools and machines. It then records them in a database that is typically accessible to the manufacturer’s maintenance, repair and operations (MRO) software to manage the required maintenance intervals for key production assets. In a well-integrated system, the MES and MRO compare actual usage to prescribed maintenance intervals and alert supervisors when a maintenance operation is due.
Alerts can also flag when an upcoming job will overrun a maintenance interval, indicating that the maintenance tasks should be performed prior to starting the job. In these instances, using cycle counts to trigger alerts is a prescriptive form of maintenance.
In contrast, predictive maintenance is based on using an MES for real-time process monitoring of actual machine and tool operating parameters, such as temperature, pressure and voltage. Sensor outputs from tools and older equipment or direct outputs from newer smart machines are recorded and monitored in real time for deviations from expected norms. When deviations are detected, supervisors are alerted, so they can intervene before problems arise.
Despite the advances offered by MES and MRO software, predictive maintenance based on process monitoring alone is a tricky proposition. In many cases, it predicts a failure after a job has been initiated. For example, process monitoring may identify signs of a pending machine or tool failure, which will trigger the need for attention.
Results like these are predictive in the sense that the worst-case scenario is avoided. But it can’t really be considered predictive in the sense that the job was allowed to start in the first place.
Instead, the optimal strategy is to adopt maintenance rules that combine cycle counts (captured with production monitoring) and process monitoring.
Maintenance based on planned versus actual cycle counts is considered preventive maintenance in that it plans to prevent problems well before they happen. Meanwhile, process monitoring is predictive in nature, but it typically occurs in a time frame much closer to when an actual failure could occur. When combined, these approaches effectively serve as a safety net for each other.
The importance of the historian database
One of the most critical aspects of MES in general — and maintenance management in particular — is the historian database, which contains time-series data points that are each tied back to specific machines, sensors and jobs. It acts in real time, first recording signals within seconds following their occurrence and then enabling analytics to be performed on the data shortly after being recorded in the historian database.
Analytics performed on historian data in the realm of predictive maintenance typically include one or more of the following four processes:
- High-low boundary conditions that trigger an alert or action when a parameter crosses a boundary threshold.
- Trend analysis to track parameters as vectors and initiate alerts when a vector is approaching a threshold.
- Pattern matching is chiefly useful where jobs are run frequently or for long periods. Current parameter patterns are compared with previous parameter patterns that have led to failures. When current patterns and previous patterns align, alerts are initiated.
- Sum of the evidence is a technique that looks at several parameters and detects situations where several variables are trending away from normal values at the same time. This method determines when a preponderance of evidence indicates that some combination of factors is likely to cause a failure soon. It is particularly amenable to the use of basic artificial intelligence (AI) tools.
Steve Bieszczat is chief marketing officer at DelmiaWorks.
