IIoT series, Part 1: Five ways to use cloud and IIoT to improve productivity: Your questions answered
Webcast presenters Alan Griffiths and Mohamed (Mo) Abuali, Ph.D. answered additional questions about topics such as augmented reality, 5G technology, and predictive analytics.
On April 18, CFE Media presented the webcast “Asset Conditioning Monitoring: Best Practices to Maximize Benefits,” sponsored by Emerson, Epicor, Infor, and Oracle + Netsuite. Alan Griffiths, principal consultant, Cambashi; Mohamed (Mo) Abuali, Ph.D., founder and CEO, IoTco, presented the webcast; the archived presentation can be found here. Webcast attendees had more questions on the topic than could be responded to during the presentation; the presenters have responded to those questions here:
Question: Would you comment on the implementations of augmented reality (AR) with digital twins for maintenance management?
Answer: AR is a great tool for realizing and executing maintenance activities, especially when enabled with IoT and predictive analytics information flow. Take this example of an aerospace original equipment manufacturer (OEM): we were monitoring and predictive computer numerical control (CNC) machine breakdowns on their spindles and ball-screw feed axis, and the resulting machine/component health and diagnosis data were sent to an AR device worn by maintenance engineers. By looking at the machine during their plant walks, the AR devices displayed real-time machine health, predictions, and diagnosis issues to the users, allowing them to proactively identify machine issues. In some cases, the AR devices displayed work instructions on how the maintenance engineers may fix the machine to allow for a timeline and reliable fix.
Q: Is 5G technology a pre-requisite enabler for truly effective IIoT?
A: In general, connectivity is a prerequisite to IIoT. Analytics together with connectivity allow for smarter IoT-enabled factories and products. 5G would allow in the future for faster data transmissions and edge computing capabilities, and that may accelerate IIoT implementations and adoptions. At this time, wired and wireless connectivity via wifi, bluetooth, or newer approaches like LORA (low-frequency radio), are more prevalent in a manufacturing environment.
Q: What use cases justify edge computing: only low-latency or volume of data or cost?
A: Edge computing and analytics is essential, not only for data volume or cost concerns, but for overall scalability of IoT in a multi-plant multi-location environment. Edge allows for AI and machine learning algorithms to process big data within an industrial PC or IoT device that would reduce the volume and dimension of that data into a more usable and transferrable meta-data like features and health that can be sent out to the cloud or server location. In many cases, data retention of large volumes of vibration for example, may not be required. With the growing computational powers and lower-cost CPUs and memory, doing analytics and computation on the edge has allowed cost-effective scalability of IoT applications.
Q: Is there a predictive analytics template that would be applicable to HVAC equipment (blowers, motors, actuators, controls, sensors, etc.).
A: Predictive analytics templates not only encompass manufacturing assets like robots and CNCs, but extends to ancillary equipment like HVAC, compressors and chillers. Some templates are intrusive, in that you need to add sensors like vibration to the process, while others are non-intrusive, in that you can utilize existing controller data like current, voltage, and temperature. In the case of HVAC equipment, many of the predictive analytics templates are non-intrusive. For a large automotive OEM, we used current/voltage surge data from HVAC units to monitor and predict failures on the equipment, without adding any additional sensors to the equipment.
Q: What kind of unsupervised training do you recommend for the factory or in the field?
A: There are many unsupervised learning techniques where only baseline data is available. For many new machines for example, where historical data is not available, unsupervised analytics models allow users to build a baseline and trend the data to monitor deviations and drift from the baseline. Once failures occur in the process, a predictive analytics solution would AI to self-learning the fault signature and diagnose it in the future, in this case becoming a supervised learning approach.