(Images courtesy: Erbessd Instruments; design Steve Thurston 2022)
Data from the circular monitors attached to the machine (the orange circles) create data that can be tracked. That data can be turned into an actionable "bad actor" list.
Two water pumps, same make and model, are not actually the same.
One could have a flaw from the factory or have been installed improperly, has a defective part or was maintained improperly. Which one is bad, or failing, is impossible to tell with the naked eye, but Erbessd Instruments in downtown Glens Falls says they have developed technology that can monitor the pumps, tell which is breaking or wearing out, and either take action, alert maintenance workers to the trouble, or both.
A factory floor might have hundreds or thousands of machines, and traditionally a handful of people check them daily or weekly, taking measurements from them to see if they are stressed, or examining them for wear, said David Howard. He is the CEO of Erbessd (pronounced: air-best).
“Our phantom wireless sensors combined with the ability to create machine learning models, takes it [monitoring] to a-whole-nother level,” David Howard said.
The monitors that Erbessd has been making can measure the amount of “wobble” or “jump” in a machine as it operates. They measure temperature and speed of rotation, amperage use and other variables also.
In their new system, when the monitors are first installed, the data is taken from the machine and sent to the computers to create a normal operating range for that particular machine, Howard said. An indenticcal machine next to it will have a different dataset. Technicians or others on the factory floor can influence what is measured and sent to the database. Howard called it “semi-supervised machine learning.”
From that dataset, the computers create a file similar to a sound waveform.
The monitoring computers look for variation in that waveform outside normal operation. For instance, if the computer sees too many waves per minute or waves that grow too tall, it sends an alert to maintenance workers telling them not just which machine is failing but exactly what that failure is. Their data can tell if the impeller blade or a bearing has gone bad inside a pump. They are right 86% of the time, he said.
The effect is increased efficiency, he said. Those maintenance workers are more focused.
“Let’s say you have 1,000 machines in a factory. Only 120 are ‘in alarm’,” he said, adding later, “In the past, they had to worry about a thousand machines and look at all the data to see if all 1,000 were healthy.”
Now they check their phone for the bad actor list and focus their energy there.
What’s more, the system can be set up for the monitoring computers to send instructions back to the machine to add oil, slow the machine down or take other more complex steps.
Howard said his system "can actually create a work order in the customer's maintenance system, and say, 'Hey, you have to go out and align this machine.'"