{"id":11039,"date":"2026-10-27T23:34:58","date_gmt":"2026-10-27T22:34:58","guid":{"rendered":"https:\/\/www.servomotorsadjust.com\/?p=11039"},"modified":"2026-10-28T20:37:02","modified_gmt":"2026-10-28T19:37:02","slug":"servomotor-predictive-maintenance-detecting-bearing-failure-before-breakdown","status":"publish","type":"post","link":"https:\/\/www.servomotorsadjust.com\/en\/servomotor-predictive-maintenance-detecting-bearing-failure-before-breakdown\/","title":{"rendered":"Can We Diagnose a Servomotor Before It Reaches the Repair Workshop?"},"content":{"rendered":"
The servomotor was still running.<\/p>\n
It had not stopped production. It had not yet been removed from the machine. And it had not arrived at our repair workshop.<\/p>\n
But the vibration data was already telling us that something was wrong.<\/p>\n
Normally, the process starts when the problem becomes evident:<\/p>\n
Failure \u2192 Machine stop \u2192 Motor removal \u2192 Diagnosis \u2192 Repair<\/p>\n
This real case shows how condition monitoring, vibration analysis and failure-pattern recognition allowed us to identify a developing bearing problem while the servomotor was still operating.<\/p>\n
More importantly, once the motor arrived at Servo Motors Adjust and was dismantled, the physical inspection confirmed what we had already identified from the vibration data.<\/p>\n<\/div>\n
<\/p>\n
Wireless vibration monitoring makes it possible to collect condition data from the machine without waiting for the motor to fail.<\/p>\n
But the sensor itself is only the beginning.<\/p>\n
The real value lies in understanding what the data is telling us.<\/strong><\/p>\n Instead of relying only on an overall vibration value, the analysis can include:<\/p>\n In this case, vibration measurements were obtained at several operating speeds, including approximately 635, 755, 880, 1,040 and 1,290 rpm<\/strong>. This allowed us to study how the vibration signature evolved with rotational speed instead of relying on a single measurement.<\/p>\n The important question was no longer simply:<\/p>\n Is the motor vibrating too much?<\/strong><\/p><\/blockquote>\n It became:<\/p>\n What is the vibration signature telling us about what is happening inside the servo motor?<\/strong><\/p><\/blockquote>\n<\/div>\n This distinction is fundamental.<\/p>\n An increase in vibration can tell us that the mechanical condition is changing.<\/p>\n A characteristic vibration pattern can help us investigate why it is changing<\/strong>.<\/p>\n The diagnostic evidence pointed towards bearing instability, increased internal clearance and probable cage degradation<\/strong>.<\/p>\n At approximately 755 rpm<\/strong>, the rotational frequency was around 12.58 Hz<\/strong>.<\/p>\n For the rear 6206 bearing, the calculated Fundamental Train Frequency, or FTF, was approximately 5.01 Hz<\/strong>. The measured value was approximately 5.05 Hz<\/strong>, showing very close correlation between the theoretical bearing kinematics and the measured vibration data.<\/p>\n The diagnosis was therefore not based on one isolated spectral peak.<\/p>\n It was the correlation between several independent pieces of evidence that increased confidence in the diagnosis.<\/strong><\/p>\n<\/div>\n This is where condition monitoring becomes genuinely valuable.<\/p>\n The objective is not simply to detect when a motor has become unusable.<\/p>\n The objective is to identify a developing condition early enough to make a controlled maintenance decision.<\/p>\n This meant that before dismantling the servomotor, we already had a technically supported diagnosis of the likely problem and knew where to focus the subsequent inspection.<\/p>\n Before the motor reached our workshop, we already knew where to look.<\/strong><\/p><\/blockquote>\n<\/div>\n This is probably the most important part of the case.<\/p>\n A diagnostic hypothesis becomes considerably more valuable when it can subsequently be compared with physical evidence.<\/p>\n Following the vibration analysis performed while the motor was operating, the servomotor was removed from service and sent to Servo Motors Adjust.<\/p>\n The motor was dismantled and the rear 6206 bearing<\/strong> was inspected.<\/p>\n The physical condition was consistent with what had previously been observed in the vibration data.<\/p>\n Inspection confirmed severe cage deterioration<\/strong> in the rear bearing, correlating with the vibration symptoms identified before disassembly.<\/p>\n This closed an extremely important diagnostic loop:<\/p>\n Vibration signature \u2192 Diagnosis \u2192 Disassembly \u2192 Physical confirmation<\/strong><\/p>\n The bearing problem identified while the servomotor was still operating was subsequently confirmed during physical inspection.<\/p>\n We did not have to wait for catastrophic failure to begin understanding what was happening inside the motor.<\/p>\n<\/div>\n This is where machine learning becomes particularly interesting for us.<\/p>\n Predictive maintenance should not simply mean an algorithm generating an alarm when vibration exceeds a predefined threshold.<\/p>\n The greater opportunity is failure-pattern recognition based on validated real cases<\/strong>.<\/p>\n Consider the complete information chain:<\/p>\n Motor operating<\/strong><\/p>\n \u2193<\/p>\n Wireless vibration monitoring<\/strong><\/p>\n \u2193<\/p>\n FFT \/ Envelope FFT \/ Time Waveform<\/strong><\/p>\n \u2193<\/p>\n Failure signature identified<\/strong><\/p>\n \u2193<\/p>\n Diagnostic hypothesis<\/strong><\/p>\n \u2193<\/p>\n Motor dismantled<\/strong><\/p>\n \u2193<\/p>\n Physical damage identified<\/strong><\/p>\n \u2193<\/p>\n Diagnosis confirmed<\/strong><\/p>\n \u2193<\/p>\n Validated failure pattern<\/strong><\/p>\n Every monitored motor that is subsequently inspected gives us an opportunity to connect two different sources of information:<\/p>\n What the failure looked like in the vibration data<\/strong><\/p>\n and<\/p>\n what was physically damaged inside the motor.<\/strong><\/p>\n That relationship creates reusable diagnostic knowledge.<\/p>\n Imagine that a specific combination appears repeatedly:<\/p>\n If motors displaying these signatures are subsequently dismantled and similar physical damage is repeatedly confirmed, those patterns become increasingly valuable.<\/p>\n We are no longer looking only at whether a vibration level is high or low.<\/p>\n We are building a relationship between vibration signatures and physically confirmed failure mechanisms<\/strong>.<\/p>\n This is where Machine Learning can add significant value.<\/p>\n New condition-monitoring data can be compared with previously validated failure patterns, helping engineers identify similarities between a motor currently operating in the field and motors whose actual failure mechanisms are already known.<\/p>\n The objective is not to replace engineering judgement.<\/p>\n The objective is to make previous failure knowledge reusable.<\/strong><\/p>\n Every confirmed failure can therefore help us interpret the next one.<\/p>\n Traditionally, the diagnosis of a servomotor starts when the motor arrives at the repair workshop.<\/p>\n Condition monitoring creates another possibility.<\/p>\n It allows the diagnostic process to begin while the servo motor is still operating at the customer’s facility<\/strong>.<\/p>\n When a developing vibration pattern is detected, we can investigate:<\/p>\n Potentially, by the time the servomotor arrives at Servo Motors Adjust, we already have valuable information about its condition.<\/p>\n For maintenance teams, this can also change the decision from an unexpected production problem into a planned maintenance intervention<\/strong>.<\/p>\n<\/div>\n There is another important advantage.<\/p>\n When vibration data exists before the repair, comparable measurements can be used afterwards.<\/p>\n We can then ask a very simple question:<\/p>\n Did the repair actually eliminate the mechanical signatures that were present before?<\/strong><\/p><\/blockquote>\n In the documented case, the dominant rotational vibration component was approximately 2.78 mm\/s before corrective intervention and approximately 0.12 mm\/s afterwards<\/strong>. Final balancing measurements also confirmed that the rotor met the specified ISO 21940 G2.5<\/strong> balance quality requirements.<\/p>\n The post-repair measurements therefore provide another important layer of evidence.<\/p>\n The objective is not simply:<\/p>\n Repair the motor.<\/strong><\/p>\n It becomes:<\/p>\n Diagnose \u2192 Repair \u2192 Measure \u2192 Validate.<\/strong><\/p>\n The traditional process is essentially reactive:<\/p>\n Failure \u2192 Production stop \u2192 Motor removal \u2192 Diagnosis \u2192 Repair \u2192 Return to service<\/strong><\/p>\n A condition-based approach changes the sequence:<\/p>\n Monitor \u2192 Detect \u2192 Diagnose \u2192 Plan \u2192 Repair \u2192 Validate \u2192 Learn<\/strong><\/p>\n That difference is important.<\/p>\n Instead of asking only:<\/p>\n What happened to this motor?<\/strong><\/p><\/blockquote>\n we can increasingly ask:<\/p>\n What is beginning to happen inside this motor, and can we intervene before it leads to a breakdown?<\/strong><\/p><\/blockquote>\n No.<\/p>\n And we believe this distinction is important.<\/p>\n Not every servo motor failure can be predicted with 100% certainty from vibration data alone.<\/p>\n Different failure mechanisms generate different signatures, and diagnostic confidence depends on factors including operating conditions, speed, load, sensor location, measurement quality and the availability of validated historical cases.<\/p>\n Machine Learning does not remove these engineering limitations.<\/p>\n But it can help us identify patterns, correlate new data with previous cases and make accumulated failure knowledge easier to reuse.<\/p>\n What makes this particular case significant is much more concrete:<\/p>\n A developing bearing problem was identified while the servo motor was still operating, and the diagnosis was subsequently confirmed through physical inspection after disassembly.<\/strong><\/p>\n That is not a prediction about what technology might be able to do in the future.<\/p>\n It is evidence from a real case.<\/strong><\/p>\n Servo motor repair will always require mechanical expertise, measurement, inspection and functional testing.<\/p>\n But the diagnostic process does not necessarily have to start when the motor arrives at the workshop.<\/p>\n Wireless condition monitoring provides the data.<\/p>\n Vibration analysis helps us understand the mechanical behaviour.<\/p>\n Failure-pattern recognition connects new signals with previous experience.<\/p>\n Machine Learning can help us scale that accumulated knowledge.<\/p>\n Physical inspection provides the evidence needed to confirm or correct the diagnosis.<\/p>\n And post-repair measurements allow us to verify whether the original mechanical signatures have actually been eliminated.<\/p>\n Together, these elements create a continuous process:<\/p>\n And every completed cycle gives us something valuable:<\/p>\n\n
<\/p>\nFrom Vibration Levels to a Failure Mechanism<\/strong><\/h6>\n
<\/p>\nThe Servo motor Was Still Running, but the Failure Pattern Was Already Visible<\/strong><\/h6>\n
<\/p>\nFrom Predictive Diagnosis to Physical Evidence<\/strong><\/h6>\n

<\/p>\n
<\/p>\n<\/div>\nWhy Physical Confirmation Matters for Machine Learning<\/strong><\/h6>\n
<\/strong><\/h6>\nEvery Confirmed Failure Adds Knowledge for the Next Diagnosis<\/strong><\/h6>\n
\n
<\/p>\n<\/h2>\n
Knowing What to Look for Before the Motor Arrives<\/strong><\/h6>\n
\n
<\/p>\nRepair Is Not the End of the Process<\/strong><\/h6>\n

<\/p>\nFrom Reactive Repair to Predictive Servo Motor Maintenance<\/strong><\/h6>\n
Can We Predict Every Servo Motor Failure?<\/strong><\/h6>\n
The Future of Servo Motor Repair Can Start Before the Motor Reaches the Workshop<\/strong><\/h6>\n
Monitor. Detect. Diagnose. Repair. Validate. Learn.<\/strong><\/h6>\n