RFID-based Contactless Vibration Sensing Methods And Applications

With the gradual improvement of Internet of Things infrastructure and the vigorous development of artificial intelligence technology,intelligent and secure sensing is becoming increasingly popular.In the industrial Internet of Things,vibration is a common and important phenomenon,such as engine vibration in factories,various motor rotations,and string vibrations.It is necessary to detect these vibration phenomena and obtain their vibration frequency and operational status.Traditional vibration detection methods,such as infrared laser tachometers,electromagnetic tachometers,and vibration sensors based on accelerometers,are contacttype and require the detection instruments to be attached to or placed on the vibrating equipment to detect vibration frequency.This method not only disrupts the normal operation of the equipment but also carries certain risks.Although video-based vibration detection is non-contact,it needs to be performed in the line of sight,and the detection method will fail if there is obstruction or at night.How to achieve safe,accurate,and non-intrusive vibration detection under non-line-of-sight conditions is an urgent problem that needs to be solved.To address the above problems and requirements,this paper proposes a non-contact vibration sensing method based on RFID signals,which can use commercial RFID devices to achieve contactless sensing of vibrating/rotating equipment and expand the system’s performance for sensing more equipment,higher frequencies,and longer distances.The main contents and achievements of this work are summarized as follows:1.For sensing model,we designed an RFID-based vibration sensing model and frequency recovery technology.To gain a thorough understanding of how RFID detects vibration,we established a fundamental and universal vibration sensing model.Firstly,we studied the impact of vibration/rotation on the signal propagation path,and then corresponded this effect with changes in the received signal,successfully establishing the connection between vibration frequency and received signal,i.e.,obtaining vibration frequency through the frequency domain of the phase.However,due to the low sampling rate of RFID(100Hz),it is difficult to recover vibration frequencies above 50Hz.To address this,we introduced compressive sensing technology and successfully restored high-frequency signals,enabling the detection of 100Hz vibration with only 0.14Hz error.After restoring the signal,we designed an adaptive artificial noise-assisted algorithm to successfully distinguish between vibration frequency and noise frequency.Finally,based on all the aforementioned methods,we designed the RF-Ear vibration sensing prototype system,which was evaluated with comprehensive experiments and tested in real-world scenarios,including common indoor devices such as fans and mixers,and common factory equipment such as fans and vacuum pumps,demonstrating excellent performance.2.For multi-device sensing,we design a multi-vibration source sensing and fault detection technology based on spectral fingerprints.To address the problem of multiple devices vibrating simultaneously and the inability to match the sensed frequencies with the vibrating devices,we propose a multi-source vibration sensing technique based on frequency spectrum fingerprints.When multiple devices vibrate simultaneously,the frequencies sensed by the system cannot be matched with the devices.We observed that even devices of the same model have slightly different hardware structures,which are reflected in the spectral characteristics of the signals.Based on this,we constructed frequency spectrum fingerprints of the vibrating devices,which successfully associate the frequencies with the devices.Using the same principle,we also constructed frequency spectrum fingerprints of device vibration failures,enabling the system to detect various vibration faults.Finally,through experiments,the system was able to sense four simultaneously vibrating devices with 90%accuracy using a single tag,and identify four types of vibration faults with 98%accuracy.By optimizing the frequency spectrum fingerprints,the computation delay was reduced by half with only a 2%decrease in accuracy.3.For high-frequency spectrum sensing,we design a high-frequency string vibration sensing technology based on multi-tag fusion.In order to enable the system to sense higher frequency vibrations,taking string vibration as an example,we have designed a compressed sensing technology based on multi-tag fusion.Compared to motor rotation(0～100Hz),string vibration has a higher frequency(100～1000Hz),smaller amplitude(less than lcm),and shorter duration(0.5～2s).The sampling of a single tag is insufficient to recover its frequency.To address this,we analyzed the error source of compressed sensing and optimized the measurement matrix by using multiple tags for simultaneous sampling,re-building the measurement matrix,and enriching the number of non-zero elements.This effectively increased the sampling rate and ultimately improved the performance of compressed sensing.Experiments show that with four tags,string vibration up to 880Hz can be recovered with an error of only 1.2Hz.In addition,based on this technology,we designed the RF-Recorder string vibration music perception system,and demonstrated how the system could record the chords played by a person during guitar playing.The experiments showed that the system has an accuracy of 94.4%for recognizing individual strings and 96.8%for recognizing chords.4.For long-range sensing,we design a meter-level distance vibration sensing technology based on spatial dimension noise reduction.In order to achieve longer sensing distances,we have developed a spatial dimension noise reduction technique.If the system performs noise reduction along the time dimension for the phase data of each tag,the maximum sensing distance for motor rotation(with a shaft length of 4cm)under an error of 0.2Hz is only 0.3m.Therefore,we changed our approach and utilized multitag simultaneous sampling to perform noise reduction along the spatial dimension(tag sequence)for the phase data at each time point.This increased the maximum sensing distance to 1.2m under the same conditions.We conducted a theoretical analysis:for noise reduction along the time dimension,the device is in a vibration state,and both vibration signals and noise are in a dynamic distribution.If the signal-to-noise ratio is small,the vibration signal is easily filtered out as noise.In contrast,for noise reduction along the spatial dimension,the device is in a stationary state,and only noise is in a dynamic distribution,so only noise is filtered out.Through experiments,we have demonstrated that with the use of 10 tags,noise reduction along the spatial dimension increases the average maximum sensing distance fourfold compared to noise reduction along the time dimension,exhibiting superior performance.