Successful diagnosis of a pump fault using sound and AI

Neuron Soundware is sometimes compared to conventional vibro diagnostics. Our experience with the oil pump for frying traditional Czech potato chips proved that our diagnostics technology using sound and AI is more sensitive and able to alert staff to potential problems in the machinery earlier than conventional methods.

The company has problems multiple times a year with the pump frame twisting. The pump frame has screw-on feet for a fully adjustable height, but from time to time they get loose and twist the frame. This leaves the pump misaligned, eventually destroying it and causing delays in production.

The production manager wants to avoid these delays, because every pump change puts production back for up to a day, with financial losses of thousands of euros. We set up both technologies, conventional vibrodiagnostics and Neuron Soundware diagnostics using sound and AI, in order to find a recommendation for which diagnostic method the potato chip company owner should use for this type of machine.

How did this comparison of conventional vibrodiagnostics and innovative AI-supported sound diagnostics turn out?

  • April 14, 2021 – we installed measuring points and took initial measurements for the equipment in normal operations (NSW + vibro)
  • April 27, 2021 at 8:45 p.m. – we took new vibration and sound readings before testing loosened screw on the machine foot
  • April 27, 2021 at 9:00 p.m. – we loosened the screw first by 90°, then by 180°, and finally by 420°
  • At every step we took vibrodiagnostics measurements and continually collected data via online Neuron Soundware AI-assisted sound monitoring
AI diagnosis of a faulty pump
The images above show clearly that Neuron soundware responded at the first 90° turn of the screw and reported an anomaly. Conventional vibrodiagnostics, in compliance with the usual ISO 10816-3, would have waited significantly longer to send the alert.

The vibrodiagnostics system just barely reached the warning level in the second phase of the test, when the frame was already significantly twisted and the pump seal was already starting to leak.

In the third, extreme phase of testing, vibrodiagnostics also showed a value only slightly above the alarm threshold, even though the pump seal was leaking practically everywhere and the elastomeric coupling member had already started significantly degrading.

We hardly need to give a long explanation of the major advantage is the speed of transmitting this information in particular. While vibration analysis is performed by an expert with set working hours, Neuron Soundware works online through HW, SW and AI continuously and remotely.

Pump anomaly detection
Recording of sound at anomaly level 0.5 as compared to the nominal state, when Neuron soundware begins sending an alert (notification) of an impending fault.
vibrodiagnostic graph
Diagnostic methods differences: Traditional vibro diagnostics, according to ISO 10816-3 standard, at 4,5mm/s notified customer and at 7,1mm/s generated alarm status. Neuron soundware sound & AI diagnostics recognized anomaly immediately after 90° screw loose and proved it’s sensitivity to anomalous machine behavior.

How Neuron Soundware works and why it can detect machine changes so quickly

The technology works on proven principles of technical diagnostics aided by modern AI and machine learning methods. Once the sensors have been placed on the machine, the system records initial data on machine operations and saves it as normal (nominal).

Preset data from the “machine library” is then added to the nominal data for training purposes. The system then compares this data with the data measured during machine operation. If the sound of the machine significantly deviates from its nominal state or the sound matches any of the anomalous sounds from the machine library, the Neuron Soundware system reports an anomaly. In other words, it sends a notification that something unusual is happening with the machine and that a technician should go and check or repair the affected part of the machine.

What is the most effective way to prevent faults on industrial machines?

The key is to install IoT equipment and sensors on a critical machine while it is still running properly. This ensures that the data collected will truly reflect the usual values for operation of that machine. We call this training the AI model and it typically goes on for several days. The model is then deployed immediately on the selected machine, adding technical knowledge of the specific machine to the continuous monitoring. The alerts allow staff to respond quickly and decide on the appropriate response to minimize losses. Most alerts allow maintenance staff to plan their operations to avoid unplanned restrictions in production.

Starting immediately, when the monitoring solution is deployed on the machine, the customer receives the service of remote equipment monitoring. They know whether the machine is up and running or experiencing a failure, and they have access to data for their own analyses, inspections and assessments of machine operations. After deploying the trained model, the customer receives information on machine anomalies including recommended steps to verify the impact on machine operations or to repair the problem. Over time, these recommendations will be more detailed and precise, even without validation from an expert human diagnostician.

The service’s value to the customer grows over time. The more faults it registers, the better the AI gets at distinguishing between faults and giving staff more precise information on the defect and more time to plan inspections or repairs.

Production directors who make this investment have a clear advantage over others, as they can make better use of their expert staff, respond more quickly to nip machine problems in the bud, and create data and knowledge bases for future advanced AI models.

Since fewer experts are available on the market and pressure on innovation and digitalization is growing, this is the time to invest in AI-assisted monitoring