Hidden in
Plain Sight

When collecting vibration data from a slowly-rotating piece of equipment, such as a mill, roller or a wind turbine main rotor, a typical MEMS sensor (with a maximum frequency of 3200 Hz) collects fewer samples per second than the industry standard Piezo accelerometer.

Many condition monitoring experts claim that this lower sample rate is sufficient for monitoring the health of the machine, and that no critical information will be missed.

But this is not strictly true. Although a lower sampling rate requires less data storage (less data is collected), meaning that samples can be collected over a longer period to improve resolution, our experts have shown that critical information about machine damage can in fact be overlooked when a lower sample rate is applied. This becomes clear through the application of envelope analysis.

Within a normal signal, there are signal components arising from the machine’s rotation, and other signal components resulting from impacts, particularly when a defect is present. An impact is a broad band excitation of the structure and does not show a particular frequency. However, there will be a short “ring” of the structure at the natural frequency after the impact.

A defect will produce impacts at regular intervals, which will eventually lead to peaks in the Amplitude spectrum. However, the initial amplitude is so much lower than that of the rotation-related frequencies, that the symptom is usually invisible.

Envelope analysis is used to identify these invisible fault symptoms by removing the signal components related to the running speed of the machine. Collected data is filtered and “enveloped” to detect repetitive impacts based on the high frequency part on the signal. The process of envelope analysis is detailed below.

Envelope analysis can therefore be used to identify recurring, high-frequency impact signals that are otherwise hidden within the machine’s normal signal. Impacts caused by bearing damage, for example, may have a low basic-repeat frequency, but each individual impact is actually a high-frequency signal containing the structure’s natural frequencies.

These natural frequencies act as an amplifier, enhancing the presence of repeating patterns and modulation within the signal, with the effects of bodily motion of the machine filtered out during analysis. Missing small traces in these higher frequency excitations, such as through the application of a lower sample rate, can seriously limit the capability of a condition monitoring system to identify faults at an early stage, especially for the reliable monitoring of very slow-running bearings.

The primary goal of a condition monitoring system is to identify faults well before they lead to significant damage or downtime. So, missing a fault at the early stage is potentially a very costly mistake.

Bachmann experts have used a study of downsampled wind turbine signals to show that the full frequency range is necessary to capture all useful information necessary for early fault identification. And this information can only be acquired with sampling rates high enough to capture the signature of regular impacts well above the normal running speed of the machine, preferably within the multi-kHz natural frequency range of the component and housing.

Although the frequency range shown in the envelope signal is normally below 1 kHz, it is normally constructed from data within the 2-10 kHz range. Considering the requirements for Anti-alias filtering, this requires a sampling rate of at least 20 kHz. MEMS sensors have a limited frequency response, both due to the mechanism that detects vibration, the limitations of the transmission path into the sensor and the filter characteristics of the sensor itself. Sufficient quality signals with high enough sample rates are therefore more readily acquired with Piezo accelerometers.

The only downside is that a higher volume of collected data requires a higher volume of storage. Fortunately, the storage capacity of modern systems means that this is no longer a significant problem. Older systems can avoid this problem by discarding the raw time signal after the enveloping has taken place.

Download our white paper: