AI-based fault detection on air compressors
Neuron soundware solution for machine health monitoring successfully detected faults on air compressors that would be difficult to detect by other diagnostic methods. This saved the production of car rims for a 50-million per year car wheel producer.
The challenge: eliminate air compressor outages
With 23 manufacturing plants in 14 countries across the globe and more than 50 million wheels produced per year, this customer is a market leader. The requirement for fluency of production and elimination of sudden outages is key for the customer.
To increase production and maintenance efficiency the manufacturer implemented a Neuron soundware solution for machine health monitoring of manufacturing air supplying compressors (critical assets) that were yet maintained using the approach “run to a failure”.
Thanks to the Neuron soundware solution the maintenance team has the information about the machine health online and in case of any change the team gets warned when the machine condition deteriorates. The fact that the system works well was confirmed by early warnings twice in a row and thus we managed to prevent the machines from shutting down and costly repairs. In addition, these failures, due to their nature, would be very difficult to detect by other diagnostic methods.
Fig. 1: Actual nGuard hardware installation at the car wheel production plant
The situation: compressor outages impact production and reputation
The customer wanted to eliminate repeating and unexpected outages of air compressors which frequently led to unplanned production downtime.
Early detection of incoming failure is very beneficial for the company as till now the maintenance team waited for the compressor to fail and switched for the spare one. This approach is called “run to failure” and by default creates a spectrum of extra cost and inefficiencies.
To protect the uptime of the critical assets, customers need to understand machine health in real-time including receipt of alert once the machine condition starts to worsen. In that way maintenance and repair can be planned in advance, unnecessary breaks in production are eliminated and machine park usage is optimized as regards primary (eg downtime cost) or secondary (eg end-customer reputation cost) impact.
Fig. 2: Air compressor deployment at the car wheel production plant
The solution: AI-based remote monitoring
Reflecting the customer need, Neuron soundware implemented the sound-based predictive maintenance solution on five chosen air compressors in one of the US plants. For each compressor, there were two sensors installed for collecting the machine sound data, one on the motor and one on the pump. These sensors have been connected to the nEdge (IIoT device), where the data is stored and processed.
The complex machine learning algorithm in nEdge has been analyzing the collected data and raising a real-time alert upon detection of anomalous behavior compared to compressors´ nominal. In the case of regular behavior, the online report has been showing standard performance in detail, confirming the OK status of the respective compressor. nEdge has been connected to the internet, and all the compressors have been monitored remotely through the Neuron soundware cloud portal.
The project started, as per common practice, with the recording of nominal sound and vibration data. Soon after, the end-to-end anomaly detection service was activated.
Despite the fact that the solution was showing nominal operation values (=confirming compressor health) most of the time, during the first 6 months, all (2) failures that occurred were detected and reported as anomalies by NSW solutions.
- Oil separator and control valve malfunction
- Processor fault in the main controller
If not detected on time, both these anomalies would have most likely resulted in critical failures, unscheduled downtime, and costly repairs. Thanks to our solution´s high sensitivity resulting in early alerts, the customer maintenance team resolved both anomalies in scheduled maintenance windows without additional downtime in manufacturing lines.
May 27th 2021 – Anomaly on compressors W2, W3 (interconnected compressors)
Fig. 3: nGuard dashboard showing the timeline of the failure detection
Maintenance Action: Defect identified on W3 oil separator with “oil overfill manifestation”, resulting in NSW immediate alert. Due to interconnection, both compressors have been checked. The control valve was replaced to reduce unwanted vibration while unloading and the pressure set-up has been corrected. Since the repair, both compressors are working fine.
Benefits: Standard offline vibrodiagnostics would not capture the issue – this type of issue typically develops within 12 hours whereas standard vibrodiagnostics is typically applied 1x / month. Neuron soundware measures continuously hence easily detected the issue as such but also its trend/escalation. Furthermore, the development of this particular failure is represented by changing frequency – moving from mid-range frequencies up to high frequencies (with mid-range frequencies gradually disappearing). Basic diagnostics (eg. ISO 10816) would be very unlikely to capture the issue, possibly leading to a catastrophic failure. Advanced diagnostics of Neuron soundware has captured the issue based on frequency of data collection, fine and functional algorithm as well as broad-spectrum sensors.
July 14th2021 – Anomaly on compressor W1
Maintenance Action: Early stage of slipping bearing has been identified as a probable root cause.
Benefits: Due to the nature of this anomaly, the traditional vibrodiagnostics could not identify the issue, as the sound fluctuates between nominal and beyond nominal continuously. With the complex signal changes over time, even an advanced vibrodiagnostician would find it extremely difficult to interpret the issue. Our nGuard system has been able to capture the issue very early and provide the customer with root cause insight, hence providing actionable information very early on to plan the maintenance and delivering significant savings.
Conclusion: nGuard enables cost-effective maintenance
Neuron soundware system of sound sensors and AI-based failure detection algorithm helped prevent major outages in Sedalia plant. This solution has also proven to be the perfect tool for service teams to monitor the condition of machinery and detect issues beyond the reach of standard methods.
With an easy-installed IoT device, the customer received real-time asset monitoring of an air compressor that helped to send an early warning of a failure to the operator. Thanks to this, the maintenance specialist was able to prioritize the assets inspection and check on the friendly visual dashboard the status of the machine from any device.
At the company level, such a solution means higher efficiency of employee time, minimized costs associated with failures, and asset life extension.
“The installation of the Neuron soundware system to analyze the air compressor anomalies has been a great benefit, from the install to production service. Monitoring the dashboards and receiving the notification when we get an anomaly is a great help in tracking issues to when they occur as well as a predictive measure to perform maintenance beforehand instead of the costly downtime.”
“The Neuron team has done outstanding work in coaching our staff on the various aspects of their equipment; determining what was an anomaly, how to interpret the alerts, as well as the basic functionality of the dashboard, and differences between using the graphs and sound wave signals.”
“This service is a vital tool that will be very helpful in the future for us, the cost savings of using the predictive analytics and alert detection alone will more than pay for the cost of the service not to mention the costly rental fees and downtime that is accrued when compressors go down.”