Author: Chris Murphy, ADI Application Engineer

abstract

This paper describes how the latest development of MEMS technology pushes the acceleration sensor to the forefront and competes with the piezoelectric sensor in the application of condition monitoring; We will also discuss how to use the new development platform that makes this possible.

Introduction to condition monitoring (CBM) and predictive maintenance (PDM)

Condition monitoring (CBM) involves using sensors to measure the current health status to monitor a machine or asset. Predictive maintenance (PDM) requires a combination of CBM, machine learning, analysis and other technologies to predict future asset maintenance cycles or possible failures. It is expected that global equipment health monitoring will develop significantly, so it is imperative to know and understand key trends. More and more CBM companies begin to adopt PDM to improve the differentiation advantage of their products. With regard to CBM, maintenance and equipment management personnel now have new options, such as wireless devices and low-cost high-performance devices. Although the infrastructure of most CBM systems remains unchanged, we can now directly integrate new MEMS technologies into systems that used to mainly use piezoelectric sensors or were not monitored due to cost barriers.

Condition monitoring – engineering challenges and design decisions

In the design of typical CBM signal chain, many different engineering specifications and technologies need to be considered. These specifications and technologies are constantly improved and their complexity is also increasing. There are various types of customers who may have expertise in a certain field, such as algorithm development (software only) or hardware design (hardware only), but they are not always proficient in both.



For developers who want to focus on algorithm development, they require that the data information base can accurately predict asset failures and downtime. They do not want to design hardware or solve data integrity problems; Instead, you want to use really high fidelity data. Similarly, for hardware engineers who want to improve system reliability or reduce costs, they need a solution that can easily connect to the existing infrastructure, so that they can benchmark the existing solution. They need to access data in a readable format that is easy to use and export to avoid wasting time evaluating performance.

Many of the system-level challenges can be solved with a platform approach—from the sensor all the way to algorithmic development—enabling all customer types.



Many system level challenges can be addressed using a platform approach (from sensors to algorithm development) to support all types of customers.

What is cn0549? How does it help extend the life of the device?

Cn0549 CBM development platform

Cn0549 status monitoring platform is a high-performance, off the shelf hardware and software solution, which can transfer high fidelity vibration data stream from assets to algorithm / machine learning development environment. The platform provides a tested and validated system solution for hardware experts, which can provide highly accurate data acquisition, reliable mechanical coupling with assets, and high-performance broadband vibration sensors. At the same time, all hardware design files are provided to help you easily integrate into the designed products. Cn0549 is also very attractive to software experts. It outlines the hardware challenges of the status monitoring signal chain, enabling software teams and data experts to directly start developing machine learning algorithms. Key features and benefits include:

U easy to install into assets while maintaining the integrity of mechanical coupling signals

U broadband MEMS accelerometer sensor with IEPE data output format

Uiepe, high fidelity data acquisition (DAQ) solution with analog input bandwidth from DC to 54 kHz

U embedded gateway captures and stores raw data for local or networked processing

U IIO oscilloscope application using ADI to display frequency domain data in real time

U directly stream sensor data to popular data analysis tools such as Python and MATLAB ®



The CBM development platform is mainly composed of four different components (as shown in Figure 1). We will introduce them one by one, and then introduce the whole composite solution.



Figure 1 Components that make up the CBM development platform

Highly accurate and high fidelity data capture and processing

Faults such as bearing problems, cavitation and gear meshing can be detected earlier when the bandwidth is wider and the sensor noise is lower. The electronic equipment for data acquisition must ensure that the measured vibration data is highly fidelity, which is very important; Otherwise, important fault information may be lost. Ensure that the vibration data is true, so that we can quickly identify trends and confidently provide predictive maintenance recommendations, thereby reducing unnecessary wear of mechanical components and extending the service life of assets.

A cost-effective way to monitor the status of less important assets

Piezoelectric accelerometer is the highest performance vibration sensor used in the most critical assets. For these assets, performance is more important than cost. For a long time, the high cost of piezoelectric sensors has hindered the implementation of condition monitoring for less important assets. Now, MEMS vibration sensors are no inferior to piezoelectric sensors in terms of noise, bandwidth and g-range, which enables maintenance and equipment managers to have a deeper understanding of less important assets that previously used troubleshooting or passive maintenance plans. This is mainly because MEMS has high performance and low cost. Now, we can use cost-effective methods to continuously monitor assets of medium and low importance. We can use advanced vibration sensing technology to easily identify and repair unnecessary wear on assets and help extend the service life of assets. This also helps to improve the overall efficiency of the equipment and reduce machine or process downtime.

Monitoring assets – detecting problems

For CBM and PDM, many different types of detection modes can be used. Most applications involve current detection, electromagnetic detection, flow monitoring and several other modes. Vibration detection is the most commonly used mode in CBM, and piezoelectric accelerometer is the most commonly used vibration sensor. In this section, we will review how technological advances have driven the development of the field of vibration sensors and what impact this has had on application decisions.

MEMS and piezoelectric accelerometer

Piezoelectric accelerometer is a sensor with very high performance, but to achieve this performance, many design choices need to be made. For example, piezoelectric accelerometers are usually used in wired installations because they consume too much power, may be very large (especially triaxial sensors), and are expensive. Considering all the above factors, it is unrealistic to use piezoelectric sensors in the whole plant, so they are generally only used for key assets.



MEMS accelerometers have not had enough bandwidth, and the noise is too high. The g-range only supports the monitoring of less important assets. This situation has not changed until recently. Recent advances in MEMS technology have overcome these limitations, enabling MEMS vibration sensors to monitor low-end assets as well as very important assets. Table 1 shows the important characteristics required for piezoelectric sensors and MEMS sensors in CBM applications. MEMS accelerometers are becoming the first choice for many CBM applications due to their small size, battery powered operation for several years, low cost and performance comparable to piezoelectric sensors.



Cn0549 CBM development platform is compatible with MEMS and IEPE piezoelectric accelerometers, and can perform benchmark comparison between different sensor types.

Table 1 MEMS and piezoelectric accelerometer

piezoelectricity MEMS
DC response
Impact resistance
Integration opportunities (3-axis, ADC, alarm, FFT)
Performance versus time and temperature
power waste
Volume (the smaller the better)
Self test
Cost of achieving similar performance
Noise
bandwidth
Mechanical connection
Industry standard interface
G range


Use of MEMS accelerometers in existing IEPE infrastructure



As shown in Table 1, MEMS accelerometers can now provide competitive specifications and performance compared with piezoelectric sensors, but can they replace existing piezoelectric sensors? In order to facilitate designers to evaluate and use MEMS accelerometers to replace piezoelectric accelerometers, ADI has designed an interface that is compatible with IEPE standard piezoelectric sensor interfaces actually used in CBM applications.

IEPE sensor interface and mechanical installation (cn0532)

Cn0532 (as shown in Figure 2) is an IEPE conversion circuit that allows MEMS accelerometers to be seamlessly connected to IEPE infrastructure like existing IEPE sensors.

Figure 2 Cn0532 MEMS IEPE conversion circuit


Uniaxial MEMS sensors usually have three output lines: power supply, grounding and acceleration output. The IEPE infrastructure requires only two lines: one line is grounded and the other transmits power / signals. The current is transmitted to the sensor. When the sensor detects vibration, the voltage is output by the same line.

Figure 3 Simplified schematic illustrating how MEMS sensors connect to existing IEPE infrastructure (power and data)

The cn0532 PCB is designed with a thickness of 90 mils to maintain the frequency response performance of the MEMS accelerometer given in the data book. The test device is installed with screws and can be tested after unpacking. Mounting blocks, PCBs and solder pastes are extensively characterized to ensure full bandwidth mechanical conversion function, maximize the visibility of various faults within the sensor bandwidth, and extend the service life of assets by capturing these faults. These solutions enable CBM designers to easily integrate MEMS accelerometers into their assets and seamlessly connect with existing piezoelectric infrastructure.



For high frequency vibration testing, the integrity of mechanical signal path is very important. In other words, from the signal source to the sensor, the vibration signal must not be attenuated (due to damping) or amplified (due to resonance). As shown in Figure 4, an aluminum mounting block (eval-xlmount1), four screw mounts and a thick PCB ensure a flat mechanical response to the target frequency range. The IEPE reference design allows designers to easily use MEMS sensors instead of piezoelectric sensors.

Figure 4 Vibration measurement test device: use eval-xlmount1 aluminum mounting block to connect eval-cn0532-ebz plate to the vibration table

Figure 5 Comparison of frequency response between eval-cn0532 and adxl1002 data manual


Vibration to bit integrity of data conversion

Now, we know that MEMS sensors can be used instead of IEPE piezoelectric sensors. They also know how to easily install them on assets while maintaining the performance given by the data table. For the CBM development platform, the important point is that it can collect high-quality conversion data (whether based on MEMS or piezoelectric sensors), and then transfer the data to the correct environment. Next, let’s look at how to acquire IEPE sensor data and maintain the highest data fidelity to develop the best CBM algorithm or machine learning algorithm. Our other CBM reference design cn0540 can help achieve the above goals.

High fidelity 24 bit data acquisition system for IEPE sensor (cn0540)

Figure 6 shows a laboratory tested and validated IEPE DAQ signal chain. This reference design provides an optimized analog signal chain compatible with MEMS and piezoelectric accelerometers. ADI not only focuses on MEMS accelerometer based solutions. Please note that the piezoelectric accelerometer provides excellent performance and is a widely used vibration sensor; Therefore, piezoelectric accelerometer is an important sensor for precision signal chain products.



The circuit shown in Figure 6 is a sensor to bit (data acquisition) signal chain suitable for IEPE sensor, which consists of current source, input protection, level conversion and attenuation stage, third-order anti aliasing filter, analog-to-digital converter (ADC) driver and full differential Sigma- Δ ADC composition. When using piezoelectric accelerometer, CBM system designers need to use high-performance analog signal chain to achieve vibration data fidelity. Designers can evaluate the performance of the signal chain by simply connecting the IEPE sensor or cn0532 IEPE sensor directly to the cn0540 DAQ reference design. ADI has extensively tested this design and provided open source design documents (schematic diagram, layout file, bill of materials, etc.) to easily integrate it into the terminal solution.



Cn0540 IEPE data acquisition board is a tested and verified analog signal chain, which is specially used to obtain the vibration data of IEPE sensor and has a signal-to-noise ratio (SNR) better than 100 dB. Most of the solutions connected with piezoelectric sensors in the market adopt AC coupling and do not have DC and sub Hertz measurement capabilities. Cn0540 is applicable to DC coupling application scenarios, in which the DC component of the signal must be retained, or the system response must be ensured to be as low as 1 Hz or lower.



Two MEMS sensors and three piezoelectric sensors are used to test the high-precision data acquisition reference design, as shown in Table 2. As can be seen from the table, the g-range, noise density and bandwidth of each sensor vary greatly, as do the prices. It is worth noting that the piezoelectric sensor still has the best noise performance and vibration bandwidth.



For cn0540, the system bandwidth is set to 54 kHz, and the signal chain noise performance is for sensors that can reach a dynamic range of >100 DB within this bandwidth range. For example, piezotronics PCB 621b40 accelerometer can reach 105 dB at 30 kHz. Cn0540 is designed to provide bandwidth and accuracy that exceed the performance of current vibration sensors, ensuring that it will not become an obstacle in collecting high-performance vibration data. It is very easy to compare MEMS and piezoelectric sensors on the same system and determine the benchmark. Whether working with MEMS sensors, piezoelectric sensors or both, cn0540 can provide the best signal link solution for data acquisition and processing, so it will be designed as an embedded solution.



When we say that MEMS sensors can provide equivalent performance at a lower cost, the SNR of adxl1002 is 83 dB, but its cost is more than 10 times lower than that of piezoelectric sensors. MEMS sensors can now replace all sensors except the highest performance piezoelectric sensors at a low cost.

Figure 6 Cn0540: applicable to IEPE sensor, capable of high performance, wide bandwidth and precision data acquisition


Table 2 MEMS and piezoelectric sensors and their corresponding noise density measurements

sensor Range

(±g)
Output range

Peak to peak (V)
Linearity (%fsr) NSD

(µg/√Hz)
Flat bandwidth (kHz) Noise at flat bandwidth (µ g RMS) Dynamic range under flat bandwidth (DB)
ADXL1002 fifty four zero point one twenty-five eleven two thousand six hundred and twenty-two eighty-two point six zero
ADXL1004 five hundred four zero point two five one hundred and twenty-five twenty-four nineteen thousand three hundred and sixty-five eighty-five point three two
PCB 621B40 five hundred ten one ten thirty one thousand seven hundred and thirty-two one hundred and four point nine five
PCB 352C04 five hundred ten one four ten four hundred one hundred and eighteen point nine three
PCB 333B52 five ten one zero point four three twenty-two ninety-eight point five zero

Embedded gateway

After the DAQ signal chain obtains the high fidelity vibration data, it must process the data, view and / or send the data to the machine learning or cloud environment in real time. This is the work of the embedded gateway.

Local real-time processing of vibration data

Intel ® (de10 nano) and Xilinx ® (Cora z7-07s) supports two embedded platforms, including support for all relevant HDLs, device drivers, software packages and applications. Each platform runs embedded ADI Kuiper Linux ®, It enables you to display time domain and frequency domain data in real time, access real-time captured data through Ethernet, connect popular data analysis tools (such as MATLAB or Python), and even connect various cloud computing instances (such as AWS and azure). The embedded gateway can transmit 6.15 Mbps (256 ksps) to your selected algorithm development tool via Ethernet × 24 bits). Some key features of the embedded gateway include:

uIntel Terasic DE10-Nano

■ dual core arm ® Cortex ®- A9 MP core processor with 800 MHz neon with double precision floating point unit (FPU) ™ Framework media processing engine

■ 1 Gigabit Ethernet PHY with RJ45 connector

uDigilent Cora Z7-07S (Xilinx)

■ 667 MHz cortex-a9 processor, tightly integrated Xilinx FPGA

■ 512 MB DDR3 memory

■ USB and Ethernet connection



IIO oscilloscope (as shown in Figure 7) is a free open source application installed with ADI Kuiper Linux, which can help you quickly display time domain and frequency domain data. It is built based on the Linux IIO framework and directly connected with the Linux device driver of ADI. It can complete device configuration, device data reading and visual display in one tool.

Figure 7 IIO oscilloscope displays FFT of 5 kHz pure tone


The ADI Kuiper Linux image also supports industry standard tools such as MATLAB and python. By using the connection layer that can work with the IIO framework, IIO binding is developed to directly transmit data streams to these typical data analysis tools. Designers can use these powerful tools, combined with the IIO integration framework, to display and analyze data, develop algorithms, and perform hardware loop testing and other data processing technologies. Provide a complete example of how to transfer high-quality vibration data to a matlab or Python tool.

Predictive maintenance development using cn0549

Developing machine learning (ML) algorithms for PDM applications typically involves five steps, as shown in Figure 8. In predictive maintenance, regression model is usually used instead of classification model to predict the impending failure. The more training data input to predictive models, the better their performance. If only 10 minutes of vibration data is input, all operating characteristics may not be detected. However, if 10 hours of data is input, the detection probability will be greatly increased. If 10 days of data are collected, the performance of the model will be stronger.

Figure 8 Steps for developing PDM applications

Figure 9 Cn0549 example use case


Cn0549 provides data collection steps in an easy-to-use system in which we can transfer high-performance vibration data streams to selected machine learning environments.



MEMS IEPE sensors come with mechanical mounting blocks that seamlessly mount MEMS sensors to assets or vibration tables. Note that IEPE piezoelectric sensors can also be used in conjunction with this system, and can be easily installed in assets, vibration tables and other devices. Before transferring the data stream to the data analysis tool, the sensor installation should be verified to ensure that there is no interference resonance. This check can be done easily and in real time using the IIO oscilloscope. Once the system is ready, you can define a use case, as shown in Figure 9, for example, a motor that operates normally at 70% load. High quality vibration data streams can then be transferred to Matlab or Python based data analysis tools such as tensorflow or pytorch (and many others).



Through analysis, you can confirm the characteristics and characteristics that can define the health status of the asset. After establishing a model that can define normal operating conditions, faults can be detected or simulated. Step 4 can be repeated to determine the key characteristics that can define the fault, thereby generating the model. The prediction model can be obtained by comparing the fault data with the data of normal running motor.



The above briefly outlines the machine learning process supported by the CBM development platform. It should be noted that the platform can ensure the highest quality vibration data transmission to the machine learning environment.



Part 2 of this article will introduce the software stack, data flow and development strategy in detail, and introduce examples of using Python and MATLAB from the perspective of data experts or machine learning algorithm developers. It also provides an overview of software integration, as well as local and cloud based development options.

Leave a Reply

Your email address will not be published. Required fields are marked *