Smart sensor nodes are the “canaries in a coal mine” for the industrial IoT, sensitively monitoring industrial equipment for vibrations, electrical current and voltage changes, temperature, sound, and even ultrasound — raising the alarm about anything potentially critical or failure producing. In fact, vibration and ultrasound are two of the earliest signs of equipment malfunction, occurring long before the moment of failure. Piezoelectric vibration sensors have traditionally been used for these applications, but microelectromechanical systems (MEMS) sensors are making inroads, thanks to several benefits.
Vibration sensors
Imagine we have a motor with a vibration sensor mounted on top. Based on the kinds of vibrations, it is possible to predict what type of failure is likely to occur in the machine in the next few days or weeks. For example, a motor could be driving a fan when one of its blades cracks or breaks. The vibrations this creates will carry a certain signature.
In another case, if the motor shakes loose from its base, this will again create a vibrational signature but with a different frequency and graph shape from the cracked blade. If an external hit to the load should knock the motor and load out of alignment, a distinctive signature would be displayed. Issues in gear mesh and bearings generally create the highest frequencies in vibration data.
Piezoelectric vibration sensors have largely been the go-to technology for this purpose. But MEMS technologies are challenging their ubiquity. The idea is not so much to replace piezos one to one in their existing use cases but to use MEMS sensors for higher-volume applications and their lower cost of monitoring industrial machines. Such nodes can provide high-quality sensor data while using wireless interface and cloud connectivity to transfer the data to the end user.
MEMS sensors are definitely smaller in size, weight, and power consumption than piezoelectric sensors, and the cost level is also much lower. MEMS offer direct digital output and are best combined with digital microcontroller hosts. Other benefits include fast recovery after high-shock events, frequency responses down to DC for very slowly rotating machines, and the possibility to sense vibrations on multiple axes.
As a result of recent innovations from suppliers such as STMicroelectronics in both the hardware and software sides of the application, designers have the tools to develop battery-powered condition-monitoring sensors of a very small size and with wireless connectivity. They can be less than 1 cm wide on each side, easily attached, and, in the newest examples, use artificial intelligence to “learn” what is being measured after they have been deployed.
IIS3DWB vibration sensor
STMicroelectronics’ IIS3DWB accelerometer is an example of a MEMS sensor especially designed for vibration-monitoring applications. The main parameters of the IIS3DWB (see table below) meet the requirements of condition monitoring in industrial environments.
The IIS3DWB is a three-axis sensor with a full-scale selectable range from 2 g up 16 g, whereas motor monitoring traditionally requires full scale up to 4 g or 8 g. The interface to host microcontroller is a digital SPI bus, and in this way, the key benefit — a high bandwidth of 6.3 kHz @ –3 dB — can be achieved with a 26.7-kHz output data rate.
The noise density of the IIS3DWB is 75 µg/√Hz, which can be reduced to 60 µg/√Hz when only one axis is enabled through the device registers. The current consumption of the sensor is 1.1 mA, making it suitable for battery-powered applications.
Other sensor features are traditional to STMicroelectronics’ family of MEMS sensors: a large 3-kB FIFO buffer, programmable filters and interrupts, embedded temperature sensor, and self-test function for end-of-production line testing. Another benefit for industrial environments is an operating temperature range from –40°C up to 105°C . It also comes with the company’s 10-year-minimum longevity commitment.
Using the IIS3DWB as an example, we’ll look at the key performance indicators (KPIs) of a vibration sensor for condition-monitoring applications. We will discuss:
- Output noise
- Wide and flat measurement bandwidth of mechanical vibrations
- Sharp out of the band rolloff for no aliasing during the measurement
- Temperature stability of the sensor output
KPI 1: Output noise
It is important to catch small problems before they become big. To detect small variations in a machine’s vibration profile, the output noise level of the sensors must be very low.
Fig. 2 below shows a simplified version of the IIS3DWB filtering chain in four blocks. Starting from the left block (the mechanical sensing element of the IIS3DWB with 7-kHz resonant frequency), the analog signal goes to the analog front end with a high-speed ADC. After that, the digital signal is filtered by a low-pass filter (LPF1) with a 6.3-kHz cutoff frequency, and the last block is a composite filter, which allows additional filtering and digital functions.
As a result of the mechanical sensor characteristics, the internal filtering, and the digital processing, the sensor reaches excellent values for its output noise density — 75 µg/√Hz for the x- and y-axis and 110 µg/√Hz for the z-axis in the three-axis mode of operation. This translates to a very low noise level while sensing vibrations in all three axes.
In single-axis mode, in which always one of the three axes is enabled, the sensor changes the way in which the internal sampling and filtering operates. Therefore, the achievable sensor resolution of the single activated axis significantly improves with noise density going down to 60 µg/√Hz for the x- and y-axes and 80 µg/√Hz for the z-axis.
KPI 2: Wide and flat measurement bandwidth
Fig. 3 below shows the frequency response of the IIS3DWB at the output of the low-pass filter (LPF1), which is flat up to the 6.3-kHz cutoff frequency at –3 dB. Thanks to this flat response, no calibration or additional filtering or equalization in the application processor is needed in the final product.
KPI 3 : Sharp rolloff for no aliasing
Another important part of the frequency response characteristic is from 6.3 kHz in Fig. 3. You can see a sharp out-of-band rolloff, higher than 90 dB per decade. Also, the attenuation is higher than 70 dB for frequencies above the 26.7-kHz ODR. Lastly, and very important as well, there is a high attenuation of more than 50 dB achieved for high-frequency signals folding potentially back inside the signal bandwidth. Sharp rolloff and high attenuation out of band eliminates harmful frequency aliasing phenomena, which could affect the quality and reliability of the vibration analysis.
KPI 4: Temperature stability of the sensor output
The last important key performance indicator of a vibration sensor is its sensitivity drift over temperature. The drift of the IIS3DWB is negligible on all three sensor axes, and it stays within ±2% tolerance from the ideal linear curve over the full temperature range from –40°C to 105°C. Again, this is a benefit for condition-monitoring applications, as no additional calibration or sensitivity compensation is needed in a microcontroller.
Note: Zero-g offset drift over temperature, traditionally important for accelerometers, is not important for a vibration sensor because the DC component is not involved.
Thanks to its KPI values, the IIS3DWB sensor provides high-fidelity vibration data. Such data can be easily analyzed by a local microcontroller either in time domain or in frequency domain by running the FFT algorithm. Any anomaly in the monitored equipment can be detected in place. Alternatively, the pre-processed data can be transferred to the cloud for post-processing to benefit from big data collection.
Evaluation tools and software
Hardware (HW) and software (SW) tools are also important for implementation of condition-monitoring applications.
On the hardware side, STMicroelectronics offers the STEVAL-MKI208V1K evaluation kit. It includes a small square board that allows easy integration of the IIS3DWB sensor into a test system during the product development phase. The board hosts the IIS3DWB vibration sensor and a connector for a flat cable, which is also included in the kit as well as a DIL24 adapter board. The board is easy to attach by screw, glue, or magnet to the place where vibrations need to be measured.
Another HW tool useful for evaluation of the IIS3DWB sensor is called the STWIN (Sensor Tile Wireless Industrial Node). The STWIN development kit is delivered in a plastic blister with a rechargeable battery, plastic shell box with mounting holes, and the ST Link programming probe. The dev kit is ready to be used either out of the box in condition-monitoring applications or it can be tailored by the user during application development at a later time.
The block diagram below shows the STWIN centered around the STM32L4, STMicroelectronics’ low-power Cortex-M4 MCU, and several sensors that target industrial applications. It also includes all of the motion sensors: the IIS3DWB vibration sensor, ISM330DHCX six-axis inertial measurement unit (IMU) with machine-learning core, IIS2DH ultra-low-power accelerometer, and IIS2MDC magnetometer. On-board environmental sensors include the STTS751 temperature sensor, HTS221 relative humidity sensor, and LPS22HH barometer. There are also two MEMS microphones — the MP23ABS1 wideband microphone with analog output and ultrasound capability and the IMP34DT05 with digital output in the audible band.
Additionally, there is STMicroelectronics’ BLE communication module, the SPBTLE-1S for wireless Bluetooth communication, and USB and RS485 transceivers in case the STWIN is used in a wired industrial application. Finally, there is a footprint for integration of secure element STSAFE-100 in case it is required by the application to ensure data protection. There are several expansion connectors — all of which are protected with ESD protection to guarantee robustness of the design.
Three SW packages are available. Two of which, the STSW-STWINKT01 and the FP-IND-PREDMNT1, target sensor data logging and vibration monitoring in time and frequency domains. The third package, the X-CUBE-MEMSMIC1, includes projects for microphone data acquisition and processing as well as ultrasound data analysis using FFT.
The STSW-STWINKT01 development package focuses on high-speed sensor data logging and allows the logging of data from all the sensors on the STWIN board at their highest output data rate, including the outputs of the machine-learning core of the ISM330DHCX six-axis IMU sensor. Data are stored either on the on-board μSD card or in a PC, connected via USB interface. The data logging can be configured and triggered from a smartphone or a PC.
The FP-IND-PREDMNT1 is specifically tailored for predictive maintenance applications. It implements both the time-domain and the frequency-domain analysis of vibration signals. The time-domain analysis evaluates the status of RMS speed and acceleration peaks for every sensor axis. The frequency-domain analysis is based on the FFT algorithm for inspecting the frequency spectrum of the vibrations.
There are two examples in this package. The first example requires an STEVAL-STWINWFV1 Wi-Fi add-on board. By connecting this Wi-Fi add-on board extension to the STWIN, vibration and ultrasound data can be directly displayed at the predictive maintanance dashboard connected to the AWS cloud. This cloud service (DSH-PREDMNT) is available for evaluation of STMicroelectronics’ sensors free of charge.
The second example is the STWIN default firmware that allows similar functionality; however, the wireless data transfer is established by an on-board BLE module directly to a mobile device. The sensor data and FFT results are then presented on the mobile device, thanks to the ST BLE Sensor app, which can be installed on Android and iOS platforms.
Conclusion
In modern industrial plants, predictive maintenance offers cost savings compared with the reactive types of maintenance used in the past. New installations as well as retro-fitting to existing equipment take advantage of powerful sensors to measure and monitor the assets such as motors, fans, pumps, etc. Utilization of several types of sensors enables anomaly detection and classification to take place several weeks or even months before the actual failure would happen.
Advertisement
Learn more about STMicroelectronics
