“Traditional maintenance is generally preventive or corrective maintenance, which usually takes up a large part of the production cost. Now, the use of IIoT (Industrial Internet of Things) to monitor the health of machines helps to achieve predictive maintenance, allowing industry personnel to predict failures, thereby significantly reducing operating costs.
Traditional maintenance is generally preventive or corrective maintenance, which usually takes up a large part of the production cost. Now, the use of IIoT (Industrial Internet of Things) to monitor the health of machines helps to achieve predictive maintenance, allowing industry personnel to predict failures, thereby significantly reducing operating costs.
As industrial equipment is generally digitized and interconnected, Industry 4.0 has been realized, and it is helping to transform production tools. It is like a game changer, making the production chain more flexible and supporting the manufacture of customized products while maintaining profitability. In addition, digitization and industrial IoT connections are also beneficial for maintenance. After using sensors, especially accelerometers, you can analyze the operating status of the machine instead of replacing worn-out parts every once in a while.
Within the framework of predictive maintenance, the operator only needs to intervene when certain early warning symptoms appear. This analysis of machine health is called condition-based monitoring (CbM), which can control maintenance costs compared to a systematic maintenance system based on a usually very conservative fixed schedule. In addition to the more flexible maintenance operation plan, it can also detect problems at an early stage, allowing operators to schedule machine downtime based on this, which is obviously much better than shutting down outside the production line.
Vibration analysis: the importance of sensors
Manufacturers use a variety of parameters to determine when to start maintenance operations. These parameters include vibration, noise, temperature measurement, etc. Among the measurable physical quantities, vibration spectrum measurement can provide the most information on the root cause of problems in rotating machines (engines, generators, etc.). Abnormal vibration may be caused by ball bearing failure, shaft deviation, unbalance, excessive looseness, etc. Each problem has its own specific symptoms, such as the source of vibration from rotating machinery.
Measure vibration with accelerometer
Vibration measurement can be performed using an accelerometer placed near the monitored element. This kind of sensor can be piezoelectric or MEMS type, the latter has more advantages, not only can provide better response at low frequencies, but also small in size.
When the ball bearing fails, every time the ball touches the crack or the defective position of the inner ring or the outer ring, an impact will occur, causing vibration, and even a slight displacement of the rotating shaft. The frequency of impact is determined by the speed of rotation, as well as the number and diameter of the balls.
But this is not all! Once a malfunction occurs, the aforementioned impact sometimes produces an audible sound, that is, a shock wave, which manifests as a low-energy spectral component and a relatively high frequency, usually greater than 5 kHz, and always far exceeds the basic rotation frequency. Only a low-noise, high-bandwidth accelerometer (such as ADI’s ADXL100x) can measure the spectrum line corresponding to the first fault signal. These accelerometers can provide valuable information for problems that cannot be detected by products with lower frequency response or higher noise. As the problem worsened, the low-energy spectral components continued to increase. In the later stage, the entry-level accelerometer can detect the vibration, but at this time, solving the fault will become urgent, and the maintenance team needs to react in a very short time. In order to avoid being caught off guard, it is very important to use a low-noise, high-bandwidth accelerometer to detect an abnormality in the first place.
Figure 1. Spectrum characteristics based on problem type. The first sign of a ball bearing failure occurs in the high frequency spectrum.
In addition to the ADXL100x series accelerometers (ADXL1001/ADXL1002/ADXL1003/ADXL1004/ADXL1005), ADI also provides many other accelerometers, which are very useful for analyzing machine status. Observed in a stricter bandwidth range, ADXL35x series products (ADXL354/ADXL355/ADXL356/ADXL357) have low noise level characteristics (noise is as low as 20 μg/√Hz, bandwidth is 1500 Hz). Unlike ADXL100x series products that provide analog output, ADXL35x series products can provide digital output to simplify the interface with the microcontroller.
Entry-level products for consumer electronics, such as ADXL34x (ADXL343/ADXL344/ADXL345/ADXL346) or ultra-low power ADXL36x (ADXL362/ADXL363) accelerometers do not have enough bandwidth or noise performance to meet the requirements of high-quality predictive maintenance. These entry-level products not only limit the diagnostic capabilities of existing equipment, but also greatly limit the availability of data used to develop future diagnostic solutions.
However, they are an excellent choice for measuring machine activity, for example, to calculate operating hours and initiate maintenance when necessary-not predictive maintenance, but preventive maintenance. These accelerometers have extremely low power consumption, so they can be powered by energy harvesters or by batteries.
If you only need to monitor and measure the sudden impact of the machine, ADXL37x series products (ADXL372/ADXL375/ADXL377) are the ideal choice. Since the impact may only change the accuracy or operating status of the machine, it may initiate (for example) corrective maintenance to correct possible problems.
From component to complete module
As mentioned earlier, the ADXL100x series have wide bandwidth and low noise characteristics. However, they use a single axis and need to be equipped with related processing electronics. In order to simplify the design, ADI provides a complete solution, using the ADcmXL3021 model to implement three-axis measurement. This 3.3 V supply voltage product includes three measurement chains based on ADXL1002, a temperature sensor, a processor and a FIFO. The whole device is enclosed in an aluminum shell (23.7 mm × 26.7 mm × 12mm), which can be installed on a rotating machine immediately. The full size of the product is ±50 g, with an extremely low noise level of only 25 μg/√Hz and a 10 kHz bandwidth. These features enable it to capture vibration characteristics in a large number of applications.
Figure 2. The ADcmXL3021 module is very suitable for implementing predictive maintenance.
Table 1. The ADcmXL3021 and ADXL100x series are very suitable for CbM applications.
The signal processing module includes not only a configurable FIR filter with 32 coefficients, but also an FFT function with 2048 nodes per axis for spectrum analysis of vibration. Then compare each frequency level of the spectrum calculated in this way with the configurable alarm threshold (6 per axis). If the spectrum components are too dense, an alert will be generated. This product can interact with the host processor through the SPI port, providing access to internal registers and a set of user-configurable functions, including advanced mathematical functions, such as calculating average, standard deviation, maximum, crest factor, and kurtosis (The fourth-order dynamic torque supports the measurement of the sharpness of vibration).
SmartMesh: suitable for IIoT networks, very suitable for the implementation of predictive maintenance
Wireless networks are particularly suitable for collecting maintenance data from vibration sensors. It does not need to be fast, but it must be robust enough to operate in an industrial environment that is usually very noisy, uses a metal structure, and has poor conductivity. It must also be able to collect data from a large number of sensors, and these sensors are not necessarily very close to the data logger.
In order to meet this demand, ADI has introduced SmartMesh® IP industrial Mesh network, which has low power consumption and high noise immunity. The last criterion is very important for the maintenance of the module. The energy harvester or lithium battery that powers it must run for 5 to 10 years, and cannot be replaced in the middle. The SmartMesh IP network is based on the 6LoWPAN standard (IEEE 802.15.4e), which is very suitable for IIoT, and is based on a proprietary protocol built around 2.4 GHz transmission. The solution includes the LTC5800 transceiver or the pre-certified LTP590x module, which is very easy to implement.
Figure 3. The SmartMesh IP network is very suitable for implementing IIoT and predictive maintenance operations.
Table 2. Cost comparison of corrective, preventive and predictive maintenance
Various technologies are used to ensure transmission reliability greater than 99.999%, including synchronization, channel hopping and time stamping, as well as dynamic reconfiguration of the Mesh network, and only the RF path is used where the signal is strongest.
Why not turn to artificial intelligence?
A variety of vibration analysis techniques currently exist. In addition to digital filtering used to overcome parasitic vibrations caused by the process itself or other components of the machine, mathematical tools can also be used to assist, such as the tools included in ADcmXL3021 (calculation of average, standard deviation, crest factor, kurtosis, etc.) ). The analysis can be done in the time domain, but the frequency analysis is the analysis that provides the most information about the anomaly and the cause of the anomaly. Frequency analysis can even be used to calculate the cepstrum that is assimilated into the frequency spectrum of the signal (inverse Fourier transform is used to calculate the logarithm of the signal Fourier transform). However, no matter which analysis method is used, the difficulty lies in determining the optimal alarm threshold so that maintenance operations are neither too early nor too late.
A method can be used to replace the traditional alarm threshold configuration, which is to introduce artificial intelligence into the fault identification process. In the machine learning phase, cloud resources are used to create representative machine models based on data from vibration sensors. After the model is created, it can be downloaded to the local processor. The use of embedded software can not only identify events that are occurring in real time, but also identify transient events so that abnormalities can be detected.
Vibration sources in rotating machines
A problem often encountered in rotating machines is the failure of the ball bearings. Performing spectrum analysis on the data obtained from the accelerometer placed near the bearing can draw many characteristic lines, amplitudes and frequencies, all of which are determined by the speed of rotation and the cause of the problem.
The characteristic frequency of the system includes:
・ Rotation frequency of bearing sleeve:
・Frequency related to defects on the outer ring (fixed):
・Frequency related to defects on the inner ring (shaft):
・ In addition to these frequency characteristics, the shock wave generated by the ball passing over the defect location (cracking, peeling, etc.) can also cause high-frequency vibration (>5 kHz), which can even be heard in some cases.
・ N: Number of balls
・ Φ: contact angle
・Faxle: the rotation frequency of the shaft
・ D: Ball diameter
・ D: Average diameter of balls
Figure 4. Ball bearings.
About the new service
In addition to building models for predictive maintenance, artificial intelligence and cloud access have opened doors to many possibilities. Correlating vibration measurement data with data from other sensors (pressure, temperature, rotation, power, etc.) can infer a lot of information about the state of the system, far more than the amount of data required for maintenance. Combining basic data can further optimize the equipment model, not only for detecting mechanical failures, but also for handling problems (for example, empty conveyor belts, pumps with no fluid flowing inside, mixers without paste, etc.). Therefore, we can consider the various services that equipment manufacturers provide to their end customers by combining equipment supply, maintenance, and statistical analysis of the performance and problems of the production line. After being equipped with a sensor module, the basic motor will become a major participant in the concept of big data.