According to MEMS Consulting, improving condition monitoring and diagnostics, and optimizing overall systems, are core challenges for today's users of machinery and technology systems. This issue is receiving increasing attention not only in the industrial sector, but in any field that uses machinery.
In the past, machines were maintained according to a schedule, and delayed maintenance could risk factory shutdowns. Today, however, process data from machines can be used to predict their remaining lifespan. In particular, records of key parameters such as temperature, noise, and vibration can be used to determine the optimal operating conditions of the machine, and even the timing of necessary maintenance. This avoids unnecessary wear and tear and allows for early detection of malfunctions and their causes. With this monitoring advantage, machines have considerable potential for optimization in terms of facility availability and effectiveness.
Predictive maintenance (PM) is based on condition-based monitoring (CBM), typically for rotating machines such as turbines, fans, pumps, and motors. CBM records the machine's operational status in real time. However, this system cannot predict potential failures or wear. Nevertheless, using PM to predict machine performance marks a turning point: with smarter sensors and more powerful communication networks and computing platforms, models can be created to detect machine changes and perform detailed calculations of their lifespan.
To create more meaningful models, we need to analyze vibration, temperature, current, and magnetic field data. Today, wired and wireless communication methods allow factories and companies to monitor equipment. Additional analytics generated by cloud-based systems provide operators and maintenance technicians with easy access to machine status data. However, local smart sensors and communication infrastructure on the machine are fundamental to these additional analyses. The following section will reveal the true nature of these sensors, along with their requirements and key characteristics.
Machine lifecycle
Perhaps the most fundamental question in machine condition monitoring is: how long can a machine run before necessary maintenance is required?
Logically, the earlier the maintenance, the better. However, to optimize operating and maintenance costs or fully realize the maximum efficiency of facility use, we urgently need experts familiar with machine performance. In motor analysis, these experts primarily come from the bearing/lubrication field, where experience shows that machine performance is the weakest area. The experts ultimately question whether the machine, relative to its actual lifespan (see Figure 1), deviates from its normal operating condition and requires repair or even replacement.
Figure 1: Machine Lifecycle
Therefore, unused machines remain in the so-called warranty phase. Problems in the early stages of the life cycle may not be ruled out, but this is relatively rare and can usually be traced back to product defects. In the later stages of interval maintenance, only specially trained service personnel can intervene in a targeted manner, including routine maintenance performed independently of the machine's condition at specific times or after specific usage cycles, such as oil changes. Therefore, the probability of failure during interval maintenance is still very low. As the machine's usage time increases, the condition monitoring phase has arrived. Therefore, future machine failures can be predicted. Figure 1 shows the following six variations in ultrasonic ranging (1); followed by vibration (2); by analyzing lubricant (3) or a slight increase in temperature (4); signs detected before failure occurs by sensing noise (5) or heat generation (6). Vibration is often used to identify machine aging. As shown in Figure 2, the vibration patterns of three identical machines during their life cycle. In the initial stage, all data are within the normal range. However, starting from the middle stage, the vibration index increases rapidly with the increase in load until the index grows to the critical range at the end of the life. Once the machine reaches its critical range, technicians must react immediately.
Figure 2: Variation of vibration parameters over time
Vibration analysis of CBM
Factors such as output speed, gear ratio, and the number of bearing components are all important for analyzing machine vibration modes. Typically, vibrations caused by the gearbox are considered multiples of the shaft speed in the frequency domain, while the characteristic frequencies of bearings are usually not considered harmonic components. Vibrations caused by turbulence and cavitation are also frequently detected, as these vibrations are often caused by air and/or liquid flows in fans and water pumps, and are therefore often considered random vibrations. They are usually stationary, and their statistical characteristics do not change. However, random vibrations can also be cyclically stationary, and therefore possess statistical characteristics. These vibrations are generated by the machine and vary periodically; for example, in an internal combustion engine, the ignition operation is generated once per cylinder.
Sensor positioning also plays a crucial role. If a single-axis sensor is used to measure predominantly linear vibrations, the sensor must be adjusted according to the direction of vibration. Although multi-axis sensors can record vibrations in all directions, single-axis sensors, due to their physical characteristics, offer advantages such as lower noise, a wider force measurement range, and a wider bandwidth.
Public requirements for vibration sensors
For vibration sensors to be more widely used in condition monitoring, two factors are crucial: low cost and small size. While piezoelectric sensors were previously more common, accelerometers based on MEMS technology are now increasingly prevalent. Accelerometers offer higher resolution, excellent drift velocity, extremely high sensitivity, and a higher signal-to-noise ratio, and can detect extremely low-frequency vibrations close to the DC range. At the same time, they are very energy-efficient, which is why accelerometers are also ideal for battery-powered wireless monitoring systems.
Compared to piezoelectric sensors, another advantage of vibration sensors is that the entire system can be integrated (System-in-Package, or SiP for short). SiP solutions are evolving into intelligent systems by integrating other important functions: analog-to-digital converters, microcontrollers in embedded firmware for specific preprocessing applications, communication protocols and universal interfaces, as well as various protection functions.
Integrated protection functionality is crucial because excessive force applied to the sensor element can damage or even render it unusable. When the integrated detection device detects a force exceeding the normal range, it will issue an alarm or directly disable the sensor element in the gyroscope by shutting down the gyroscope's internal clock, thereby protecting the sensor element. The SiP solution is shown in Figure 3.
Figure 3: MEMS system-in-package (bottom left)
As demand in the CBM field increases, so do the requirements for sensors. For practical CBM, the requirement for sensor measurement range (full-scale range, or FSR) is now partially greater than ±50g.
Because acceleration is proportional to the square of frequency, faster frequencies can achieve acceleration relatively quickly. Equation 1 proves this:
Let variable 'a' represent acceleration, 'f' represent frequency, and 'd' represent vibration amplitude. Therefore, assuming a vibration of 1 kHz, an amplitude of 1 μm would produce an acceleration of 39.5g.
Regarding noise performance, the values should be relatively low over the widest possible frequency range, from extremely low frequencies close to DC to mid-range frequencies in the double digits (kHz). Therefore, bearing noise should be detectable at very low speeds, excluding other human factors. However, this is precisely where vibration sensor manufacturers currently face significant challenges, especially for multi-axis sensors. Only a few manufacturers can provide sufficiently low-noise sensors with bandwidths greater than 2kHz suitable for multi-axis sensing.
The single-axis sensors in Analog Devices' ADXL100x series are designed for higher bandwidths. They offer bandwidths up to 24 kHz (resonant frequency = 45 kHz) and measurement ranges up to ±100 g at extremely low noise levels. Due to their high bandwidth, most faults occurring in rotating machinery, such as damaged sliding bearings, machine instability, running friction, loose parts, gear tooth defects, bearing wear, and cavitation, can be detected using this sensor series.
Feasibility analysis method for CBM
We can perform machine state analysis in CBM using various methods. The three most commonly used are probably time-domain analysis, frequency-domain analysis, and combined time-frequency analysis.
1. Time Domain Analysis
In time-domain vibration analysis, we need to consider the effective value (root mean square, or RMS for short), peak-to-peak value, and amplitude (see Figure 4).
Figure 4: Amplitude, RMS value, and peak-to-peak value of harmonic vibration signal
Peak-to-peak value reflects the maximum deflection of the motor shaft, and therefore the maximum load value can be calculated from it. In contrast, amplitude value refers to the magnitude of vibration, which can be used to identify abnormal impact events. However, amplitude value does not consider the duration of the vibration event or the energy generated, nor its destructive force. Therefore, the RMS value is usually the most meaningful because it takes into account both the vibration time history and amplitude. Thus, the relevance of the RMS vibration statistical threshold can be obtained through the dependence of these parameters on motor speed.
This type of analysis proves to be very simple because it requires neither basic system knowledge nor any type of spectral analysis.
2. Frequency Domain Analysis
Through frequency domain analysis, time-varying vibration signals can be decomposed into their frequency components using a Fast Fourier Transform (FFT). The resulting amplitude and frequency spectra can be used to monitor specific frequency components, their harmonics, and sidebands, as shown in Figure 5.
Figure 5: Spectrum diagram of vibration versus frequency
Fiber Stroke Theory (FFT) is a widely used method for vibration analysis, particularly suitable for detecting bearing damage. It assigns a corresponding component to each frequency component. Because contact between rolling elements and defective areas can cause certain faults, the dominant frequency of recurring pulses from these faults can be filtered out through FFT analysis. Since their frequency components differ, different types of bearing damage (such as outer ring damage, inner ring damage, or ball bearing damage) can be distinguished. However, this analytical method still requires precise information about the bearing, motor, and the overall system.
Furthermore, the FFT analysis process requires repeatedly recording and processing discrete-time blocks of vibration within a microcontroller. Although this requires slightly more computation time than time-domain analysis, this method allows for more detailed damage analysis.
3. Time-frequency combined analysis
This type of analysis is the most comprehensive because it combines the advantages of both methods. Statistical analysis in the time domain provides information about the intensity of vibrations in the system over time, as well as whether the system is operating within permissible limits. Frequency-based analysis can monitor the velocity in the form of the fundamental frequency, as well as other harmonic components needed to accurately identify fault symptoms.
Fundamental frequency tracking is particularly crucial because the RMS value and other statistical parameters change with velocity. If the statistical parameters have changed significantly compared to the last measurement, technicians must check the fundamental frequency to prevent false alarms.
It is normal for the measured values of the three analysis methods mentioned above to change over time. Methods for monitoring the system may include first recording the machine's "health status" or generating a so-called "fingerprint" (initial data). This is then compared to the continuously recorded data. If excessive deviations occur or the corresponding thresholds are exceeded, a response is required. As shown in Figure 6, FFT may issue a warning (2) or an alarm (4). Depending on the severity, on-site intervention by technicians may also be necessary.
Figure 6: Threshold and reaction of FFT
CBM using magnetic field analysis
With the rapid development of integrated magnetometers, measuring stray magnetic fields around motors has become another effective method for monitoring the condition of rotating machinery. The measurement is non-contact, meaning there is no direct connection between the machine and the sensor. Like vibration sensors, magnetic field sensors are also available in single-axis and multi-axis configurations.
For fault detection, stray magnetic fields should be measured in both the axial (parallel to the motor shaft) and radial (perpendicular to the motor shaft) directions. The radial magnetic field force is typically weakened by the stator core and motor housing. It is also affected by the magnetic flux in the gaps. The axial magnetic field is generated by the current in the squirrel-cage rotor and the end windings of the stator. The position and orientation of the magnetometer are crucial for measuring both magnetic fields. Therefore, it is recommended that the magnetometer be installed close to the shaft or motor housing. Temperature measurement is also necessary because magnetic field strength is directly related to temperature. Therefore, most magnetic field sensors today integrate temperature sensors. However, sensor calibration (such as temperature drift compensation) should not be overlooked.
Applying FFT to motor magnetic field condition monitoring is similar to vibration measurement. However, for assessing motor condition, even frequencies in the low-frequency range of a few hertz to approximately 120 Hz are sufficient. Line frequencies are prominent, and in the event of a fault, the low-frequency components of the spectrum will dominate.
When the rotor bars in a squirrel-cage rotor are damaged, the slip value plays a decisive role. Its magnitude depends on the load; under ideal no-load conditions, its value is 0%. Under rated load, for a healthy machine, its value is between 1% and 5%, increasing accordingly when a fault occurs. Therefore, when performing CBM (Continuous Braking Measurement), measurements should be taken under the same load conditions to eliminate the influence of load.
Predictive Maintenance (PM) Status
Regardless of the type of condition monitoring, even with the most intelligent monitoring concepts, it's impossible to guarantee 100% against unexpected downtime, mechanical failures, or safety accidents. We can only mitigate these risks. Productivity management (PM) is increasingly becoming a widely discussed topic in the industrial sector, considered a prerequisite for the future sustainable development of production facilities. Therefore, we need continuous innovation to drive technological improvements and rapid development. Currently, the financial deficit mainly stems from a comparison between customer benefits and costs.
Nevertheless, many companies across various industrial sectors have recognized the importance of Product Management (PM), not only in the service sector but also for the greater opportunities it offers for future business growth. Despite significant challenges, the technological viability of PM remains substantial, particularly in data analytics. However, PM is currently driven by opportunism. Future business models are expected to be primarily determined by software components, with the value-added share of hardware gradually decreasing. In conclusion, considering the higher returns from longer machine uptime, our current investments in PM hardware and software are worthwhile.
