The proliferation of alarms in control rooms and in the field can cause “alarm fatigue” for some operators. IIoT applications for human-in-the-loop decisions often involve additional alerts and notifications. Choosing the right problem to solve is often a critical ingredient in avoiding additional alarm fatigue. Here I'll give you an example of a company that experienced alarm fatigue, and explain what you should do to set yourself up for success.
A few years ago, an industrial company decided to implement a proposed “predictive maintenance” solution on certain critical pieces of equipment. After a lot of time and money, the system was implemented. However, it wasn’t a predictive maintenance system. Instead, it raised alerts when potential “events” were occurring, and asked operators to label the events. Over time, the system would know whether an event actually indicated a potential problem.
This approach flooded operators with sensor deviations that they had to investigate, evaluate, and label. The task was so overwhelming that the company contemplated staffing a full-time center just to handle the alerts raised by this system. The deluge of system alerts overloaded already stretched operators. Eventually, the system was torn out. It never created any value for the company.
In retrospect, the company realized something else: all of the critical equipment chosen for this project was highly engineered and well-instrumented, and it rarely failed. Each sensor already had alarms attached to threshold values (“high high” and “low low” alerts). Managing these alarms already created significant work for the operators. In addition, the most critical assets were also monitored by OEMs as part of their aftermarket service packages. These notifications and suggestions also created work for the operators.
The new system – the so-called predictive maintenance system – was actually the third line of defense after rules-based alerts and OEM equipment monitoring for critical assets.
Given the expense of designing and implementing the system, and the significant time involved to operate it, much less ongoing software costs, the business case for such a system was minimal.
Operators already suffer from alarm fatigue in most complex industrial control environments. Many experienced operators already know which alerts to log, which to escalate, and when to take immediate action. Learning how to manage new alerts in a new system may be daunting for many industrial operators – and particularly unnecessary for those assets that are least likely to fail.
What will you do differently, given new information?
It’s nice to know if equipment is behaving in a way that's unexpected. However, that knowledge may or may not be important. Even if the indications are important, they may not necessarily be actionable (for instance, you might not be able to do anything about an event in time to be meaningful to the business).
Asset owners should be pragmatic
If you already have reliable equipment that rarely fails, sometimes the right decision is still to implement additional layers of monitoring and detection. This is most often true when failure could lead to catastrophic loss of life.
More often, however, the right decision is to skip the extra investment in technology and take a different approach to the risk of potential failure: buy insurance, holding more spare parts, or doing more scheduled maintenance.
If your company is serious about IIoT implementation, set yourself up for success. Focus on wins that are both valuable to the business and feasible to implement. This feasibility involves technical aspects – for instance the availability of required data – and the ability to use the solution given existing business processes, people, skills, and related systems.
More alerts aren't necessarily better
Choosing the right problem for IIoT implementation isn’t as simple as it seems. Sometimes the most apparently obvious use cases aren’t the best candidates for a successful long-term program. It’s important to think hard about all of the important pieces to an IIoT solution in conjunction:
- the user workflow, interface and experience
- the data environment, instrumentation, and solution architecture
- the analytical expectations and the related data engineering and data science requirements
- and the fundamental business process requirements – how an operator will deal with yet another notification or alarm.
Ultimately, operators don’t just need more alerts. They need clear insight into what actions to take, and why.
Whitepaper: The Reality of Predictive Equipment Maintenance - Cluster-Based Anomaly Detection
Want to learn how you can effectively implement predictive analysis? In our whitepaper, The Reality of Predictive Equipment Maintenance - Cluster-Based Anomaly Detection for Industrial Companies, we provide you with useful information on how you can bypass common challenges and effectively implement predictive analytics across your operations.
This article was originally posted June 25, 2019 and has been updated.