Predictive maintenance: Monitor, measure and act: Your questions answered
The "Predictive maintenance: Monitor, measure and act" webcast was presented live on Sept. 25, 2018, by David Auton from C&W Services. The webcast can be found here. He supplied written answers to some of those questions that weren’t addressed from the webcast attendees:
Question: What kind of tool do we need to use to establish an adequate predictive maintenance (PdM) strategy?
Answer: My recommendation is to start with a reasonable service strategy, define the specific deliverables that will be needed to achieve the strategy and then look at what technology can provide sensory data sufficient for achieving. The technology is a tool, it is not the strategy or even capable of critical decisioning to develop a strategy. That responsibility rests with the technical team that now have more tools to make the accomplishments easier.
Q: The P-F curve is an excellent tool to define the best strategy, but how can we define the cycle for predictive if don’t use the correct monitoring technology?
A: Remember that the P-F curve is designed for analyzing the system component behavior, not the entire system behavior. Using the monitoring or PdM approach and the PF curve, we intervene when needed, not based on a calendar cycle. The monitoring should be done with sensors that provide the correct component failure indicator (temperature, vibration, amperage draw, etc.).
Q: In areas where direct access to wireless data is not available (e.g. offshore with limited satellite access and no fiber connection), how are wireless connection networks used?
A: The objective is to transfer the physical measurements obtained from the sensors to the machine learning application. In such cases, there needs to be a cost analysis of employing satellite communications vs. having the machine based application on-site. Updates to the coresite-based software would need to be done on a relevant basis but otherwise the solution should be self-contained on the campus or rig.
Q: Would you please talk a little bit about the challenges with establishing a reliable wireless network architecture and your solutions and practices to mitigate or eliminate them?
A: The closer to industry accepted models, the less expensive it will be to implement. Using existing building LAN, for example, is less expensive than adding a VPN, which is less expensive than adding a cellular network, etc. Also, wireless technology makes implementing easier, but not the only solution. Traditional hard-wired capability is more reliable for high stress environments. Few campus networks have LAN access points near the cooling towers and similar challenges must always be evaluated.
Q: Can energy consumption be a good predictor of failure?
A: Depends on the fault analysis and FEMA conclusions. Increased energy consumption could be due to equipment internal frictions or production load or other factors.
Q: In your experience, from what minimum scale of a plant will business case work?
A: How much time do you spend monitoring your equipment condition today? I have seen similar campus size with very different labor investments. How much of that labor force can be replaced with automated monitoring? A good trade-off is the example I provided where around $40K in labor can be re-directed. Basic wireless solutions can be considered at that price point.
Q: Do you focus on refrigeration coil cleaning at all for energy savings?
A: Most existing BAS can already monitor pressure drops across coils and filters (with the correct sensors installed). I would suggest that as your first avenue for a monitoring solution.
Q: With the added sensors, especially vibration, will the data collection system like a supervisory control and data acquisition (SCADA) server will have to be upgraded. Could you upgrade using the existing SCADA and tie the tags to the predictive software. Who configure all this data in the software?
A: Depending on the sensors used, the data may be transferrable across platform. I have to date only seen sensors connected to proprietary analysis applications and the outputs (alarms, warnings, values, etc.) then sent across standard communication protocols after the analysis.
Q: What about running data creating a virtual system for analysis purposes? Is this approach useful?
A: Virtual systems are used for scenario considerations. Our focus has been on the current condition of systems. Although more specific data can be collected that may help in the generation of virtual systems, that would be a separate exercise.