Maintenance With AI Detect Bearing Damage Early

Related Vendors

Altair Engineering GmbH

Wuxi InfiMotion Technology Co., Ltd.

Bearing damage, which leads to machine failures and production interruptions, causes high costs. Predictive maintenance with AI helps to predict such damage early, optimizing maintenance cycles and avoiding downtime.

For efficient machine operation, smoothly functioning bearings are absolutely essential. Since bearing damage can lead to costly repairs or even downtime for entire systems, monitoring bearings and the associated use of sensors play a crucial role in efficient system operation.

These sensors provide the necessary data for an increasingly important aspect of modern industrial operations: equipment monitoring and predictive maintenance of machines. This article explores how the data analysis and AI platform Altair RapidMiner is used to analyze a comprehensive dataset on bearing performance under various operating conditions to determine the condition of a bearing. The Altair solution enables predictive maintenance through meaningful results—thereby reducing equipment downtime.

Significantly Reduce Downtime

In an application case from the recently introduced E-Book titled "100 AI-Powered Engineering Use Cases", the task was to significantly reduce machine downtime caused by recurring bearing damages. These damages included inner and outer raceway damages, roller damages, and complete bearing failures, leading to production losses the company aimed to avoid in the future.

Condition Monitoring With IoT Sensors

Statements about the condition of the bearings were made based on sensor data from test benches. To detect emerging damages in time, efficient processing and evaluation of this real-time data is required. Since large data volumes are generated in the process, the team used artificial intelligence (AI) for the automatic collection and analysis of the data. AI is capable of identifying patterns, clusters, and anomalies in the data and providing meaningful predictions that enable informed decision-making.

Goal: Development of A Predictive Model for Real-Time Condition Monitoring

The AI platform Altair RapidMiner was used, specifically the included solution Altair Panopticon, to meet the data visualization requirements with streaming data. The real-time data visualization tool processes complex datasets from multiple data streams and can perform on-the-fly comparisons with historical time series data. With simple drag-and-drop functionality, meaningful data visualizations can be created, providing clear insights into rapidly changing data pools—and enabling very quick, informed decision-making based on these insights. If anomalies are detected in the incoming data, alerts can be generated, and emails or SMS messages can be sent to the responsible team.

The initial goal was to create a predictive model using historical acceleration sensor data from bearings, which would subsequently enable real-time condition monitoring and failure prediction.

The bearing dataset to be analyzed contained data from bearings whose behavior under various conditions was recorded using a bearing test rig. The data includes features such as vibration signals, temperature measurements, and many others, which are crucial for accurately predicting the bearing condition.

Test Rig Setup And Procedure

• For the test rig setup, four double-row, force-lubricated bearings were mounted on a shaft (see figure).
• Using an AC motor connected to the shaft via V-belts, the shaft's rotational speed was kept constant at 2000 rpm.
• A spring mechanism additionally applied a radial load of 6000 lbs to the shaft and bearings.
• Highly sensitive quartz ICP accelerometers on the bearing housing were used as sensors to capture parameters.
• For the first dataset, two accelerometers were installed on each bearing, one for capturing acceleration along the x-axis and the other for the y-axis.
• For datasets 2 and 3, one accelerometer per bearing was used.
• On the test rig, it was observed that all failures occurred only after exceeding the intended bearing lifespan of over 100 million revolutions.

How the Acceleration Was Extracted

Sensor data acquisition and logging were performed using a multifunction data acquisition card, which can output control signals from test benches and record these signals with precise timing. The data was transformed into the frequency domain using the Fast Fourier Transform. The five highest amplitudes were selected and divided into various ranges (frequency below 1200 Hz to very high frequency => 6000 Hz). The resulting acceleration was then calculated based on the values in the x- and y-directions, and all statistical parameters were extracted.

Panopticon Dashboard Visualizes Live Predictions

To train ML models for fault prediction and proactively prevent potential failures, statistical, domain-specific, and visual features were extracted from the raw measurements. Random Forest was among the methods used to find approximate solutions for complex problems and make the best possible predictions.

The preprocessed dataset is then used for model creation and evaluation. The model evaluation is performed based on evaluation criteria such as accuracy, classification error, kappa value (agreement), or the so-called "weighted mean recall." Finally, live data is streamed, real-time predictions are generated, and visualized on the Panopticon dashboard.

Successfully Reduce Downtime With AI

Panopticon captured and processed historical sensor data and, in combination with real-time measurements, provided the manufacturer with real-time assessments and precise insights into the condition of the bearings, significantly accelerating decision-making. By using AI technology, the tool minimized production downtime and outages by predicting unexpected bearing failures and identifying fault types and anomalies early on. This enables the company to perform proactive maintenance, which not only reduces downtime but also increases efficiency and lowers costs. Since failures can be predicted, this approach helps the company reduce maintenance costs and plan spare part requirements more effectively.

Future-Proof Authentication with Universal RFID Readers

Universal RFID Readers: One Solution for All Credentials