Lately there has been a lot of discussion about the differences surrounding electric motor prognostics with expert systems and machine learning IoT devices. The difference became glaringly obvious right before the show floor opened on the last day of the SMRP 30th Annual Conference in Raleigh, NC. One of the other booths had a series of electric motors and drives that they were also going to need for a fully booked precision maintenance workshop the next day. Following the failure of two of the drives that morning they identified the fault to the conference center, who took no action, and then they said they went to a machine learning test equipment manufacturer nearby, who was unable to evaluate/troubleshoot due to the limits of their equipment. So, they came several aisles over to the MotorDoc booth in which I grabbed the EMPATH and a few tools and went over to their booth.
Within a few minutes we were able to identify that there was a short in the conference center’s power block that put full voltage and current into ground. We were then able to demonstrate the condition to the conference center electricians (they understood the data) and the floor manager. In the end, the conference center offered to pay for the damaged equipment (and whatever else was negotiated) all before the final day got into full swing.
It brought up a very important concept that I’ve been presenting in papers and in person. ML/AI is explicit – it can ONLY tell you what it was trained to, bias and all. With an expert system, in addition to being able to provide immediate detection and fault identification, they are able to be flexible enough to work outside of the box, providing information that they did not have to be trained to provide. With an ML/AI system that requires training to detect specific faults, once outside their training box they lose their usefulness.
This also extends into issues associated with digital transformation and the idea that AI/ML would replace technicians. While machine learning and data science have their place in freeing up expertise from routine tasks, there are none that replace experts. Experts also require systems that are flexible enough to provide data other than that which is fixed to meet the needs of the IoT device. In the example at the beginning of the article, the ability to use the EMPATH as an oscilloscope provided us the ability to troubleshoot components that were not electric motors.
This is one of the reasons why EMPATH has been easily used to monitor incoming power in special applications in addition to data collection of power quality issues at the point of common connection. The use of simple rules developed by industry experts with experience-based alarm settings and pattern analysis in general outperforms machine learning systems that can be tricked when conditions change (more on this in the next article).
