CN-121978751-A - Artificial intelligence micro-vibration detection method based on FMCW millimeter wave radar

CN121978751ACN 121978751 ACN121978751 ACN 121978751ACN-121978751-A

Abstract

The invention is suitable for the technical field of building micro-vibration detection, and provides an artificial intelligent micro-vibration detection method based on an FMCW millimeter wave radar, which comprises the steps of adopting a self-adaptive optimization strategy to dynamically adjust the transmitting signal parameters of the FMCW radar and collecting radar intermediate frequency signals; preprocessing the acquired radar intermediate frequency signal, synchronously extracting the time-frequency domain physical characteristics and the deep learning abstract characteristics of the radar intermediate frequency signal, fusing the cross-domain characteristics to construct a fused characteristic vector, and inputting the fused characteristic vector into a light-weight transducer micro-vibration recognition model trained by a domain self-adaptive framework for processing to obtain a micro-vibration detection and classification result. According to the invention, closed-loop dynamic optimization of the emission parameters is realized through reinforcement learning, the environment is actively adapted, the signal-to-noise ratio of the micro-vibration signal is improved, a high-quality data base is laid for subsequent analysis, and the performance and the interpretability of the model are considered through fusion of physical characteristics and deep learning characteristics.

Inventors

LUO JUN
YIN WENFEI
YANG YANG
ZHANG YANG
YIN SONGLIN

Assignees

中煤第三建设（集团）有限责任公司
中煤矿建(安徽)科技有限责任公司

Dates

Publication Date: 20260505
Application Date: 20251127

Claims (8)

1. An artificial intelligence micro-vibration detection method based on an FMCW millimeter wave radar is characterized by comprising the following steps: S1, adopting a self-adaptive optimization strategy to dynamically adjust the transmitting signal parameters of the FMCW radar and collect radar intermediate frequency signals; s2, preprocessing the acquired radar intermediate frequency signal, synchronously extracting the time-frequency domain physical characteristics and the deep learning abstract characteristics of the radar intermediate frequency signal, and carrying out cross-domain characteristic fusion to construct a fusion characteristic vector; S3, inputting the fusion feature vector into a light-weight transducer micro-vibration identification model trained by a domain self-adaptive framework for processing so as to obtain a micro-vibration detection and classification result; And S4, dynamically adjusting the reasoning precision and the calculation force configuration of the micro-vibration identification model according to the signal complexity indicated by the micro-vibration detection and classification result, and realizing the balance of the detection precision and the real-time performance.
2. The artificial intelligence micro vibration detection method based on FMCW millimeter wave radar according to claim 1, wherein, In the step S1, the method includes the following steps: s11, calculating the signal-to-noise ratio of a received radar signal in real time, and taking the signal-to-noise ratio as an environmental state observation value of the reinforcement learning agent; s12, constructing a reinforcement learning model which takes the slope, bandwidth and period of a Chirp signal as an action space and the signal-to-noise ratio improvement and data effectiveness as a reward function; S13, outputting the current optimal Chirp signal modulation parameters through an interactive iterative optimization strategy network of the intelligent agent and the radar signal environment; s14, controlling the radar radio frequency front end according to the modulation parameters, and generating and transmitting FMCW signals adapting to the current environment and the target characteristics.
3. The artificial intelligence micro vibration detection method based on FMCW millimeter wave radar according to claim 2, wherein, In S2, the method includes the following steps: S21, carrying out wavelet packet transformation on the preprocessed radar signals to decompose the preprocessed radar signals into preset frequency bands, and extracting instantaneous amplitude, instantaneous frequency and energy envelope of signals of each frequency band to form physical feature vectors; S22, inputting the same preprocessed radar signals into a lightweight one-dimensional convolutional neural network to extract high-level abstract semantic features of the radar signals to form abstract feature vectors; S23, adopting a multi-head attention mechanism to interact and distribute weights between the physical feature vectors and the abstract feature vectors to obtain weighted fusion feature representation; and S24, splicing the weighted physical features and the abstract features, performing dimension reduction and integration through a full connection layer, and outputting a final fusion feature vector.
4. The artificial intelligence micro vibration detection method based on FMCW millimeter wave radar according to claim 3, wherein, In the step S3, a lightweight transducer micro-vibration recognition model trained by a domain adaptive framework is trained by the training process comprising the following steps: Collecting a large number of unlabeled generalized scene radar signal data serving as a source domain and a small number of labeled specific micro-vibration scene data serving as a target domain; Constructing a classification model which takes the fusion feature vector as input and takes a lightweight transducer as a core framework; in model training, adopting a standard classification loss function, introducing domain discriminant loss, and simultaneously promoting model learning domain invariant features through a gradient inversion layer; The classification task and the domain adaptation task are combined and optimized, so that the characteristic distribution difference between the source domain and the target domain is reduced while the discrimination capability of the model on the micro-vibration signals is maintained.
5. The artificial intelligence micro vibration detection method based on FMCW millimeter wave radar according to claim 4, wherein, In the S3, the lightweight transform core architecture comprises modeling and sequence compression of local dependency relationship of an input fusion feature vector by adopting a one-dimensional convolution layer; Wherein, the A grouping self-attention mechanism is used to replace a standard multi-head self-attention mechanism so as to reduce the computational complexity; a depth separable convolution structure is adopted in the linear transformation layer, so that the model parameter is further reduced; and a gating linear unit is introduced into the feedforward neural network, so that the nonlinear expression capacity of the model is enhanced.
6. The artificial intelligence micro vibration detection method based on FMCW millimeter wave radar according to claim 5, wherein, In the step S4, dynamically adjusting the inference accuracy and the computing power configuration of the microvibration identification model, including: according to the micro-vibration detection and classification results, calculating the time domain variance and the frequency domain entropy of the signal as evaluation indexes of the signal complexity; Presetting a complexity threshold, wherein the complexity threshold comprises a first threshold and a second threshold; when the evaluation index is lower than a first threshold value, enabling a shallow sub-network obtained through knowledge distillation to perform reasoning; When the evaluation index is between the first threshold value and the second threshold value, enabling a standard light-weight transducer model; When the evaluation index is above the second threshold, enabling an enhancement model comprising more attention heads and a wider feed forward network for reasoning.
7. The artificial intelligence micro vibration detection method based on FMCW millimeter wave radar according to claim 6, wherein, The shallow sub-network and the standard light-weight transducer model are compressed by a method combining model quantization and knowledge distillation, and the method comprises the following steps: Training phase: Taking the enhancement model as a teacher model, taking the shallow sub-network or the standard model as a student model, and carrying out knowledge migration by softening output distribution and minimizing KL divergence; deployment phase: INT8 quantization is performed on the weights and activation values of the student model to reduce storage and computation overhead.
8. The artificial intelligence micro vibration detection method based on FMCW millimeter wave radar according to claim 7, wherein, The system comprises a signal modulation module, a feature fusion module, an intelligent analysis module and a dynamic reasoning module; Wherein, the The S1 is executed by a signal modulation module; The S2 is executed through a feature fusion module; the S3 is executed through an intelligent analysis module; and S4, executing through a dynamic reasoning module.

Description

Artificial intelligence micro-vibration detection method based on FMCW millimeter wave radar Technical Field The invention belongs to the technical field of building micro-vibration detection, and particularly relates to an artificial intelligent micro-vibration detection method based on an FMCW millimeter wave radar. Background The structural health monitoring, especially for early micro deformation and micro vibration monitoring of large-scale infrastructures such as bridges, high-rise buildings and the like, is a core technical means for preventing structural disasters and guaranteeing public safety. Traditional monitoring methods, such as total stations, GPS positioning systems or fiber optic sensors, are widely used, but have significant limitations. Therefore, in recent years, the frequency modulation continuous wave millimeter wave radar technology has great potential in the micro deformation monitoring field due to the advantages of high precision, high resolution, strong anti-interference capability, all-weather operation and the like. The FMCW radar is used for accurately measuring very small displacement changes of a target by emitting continuous waves with linearly changing frequencies, receiving echoes reflected by the target and extracting intermediate frequency signals containing distance and speed information through mixing. However, the application of millimeter wave radar to long-term, high-precision monitoring of urban infrastructure still faces a serious set of technical challenges. First, in the hardware level, in order to achieve reliable acquisition of long-distance, weak signals, the radar antenna needs to have high gain and high isolation. Array antennas in the prior art often have the problem that the side lobe level is too high, so that the radar is easy to receive interference signals from the side lobe, and the signal to noise ratio and accuracy of monitoring data are seriously affected. Secondly, in order to meet the convenience and environmental adaptability of on-site deployment, the radar system needs to develop toward miniaturization and integration. But introduces dense signal integrity, power integrity, and electromagnetic compatibility issues in the high frequency millimeter wave band. Finally, high-density integration of multiple devices such as an internal processor, a memory, a radio frequency front end and the like in the system-in-package is easy to generate serious mutual interference. Disclosure of Invention The embodiment of the invention aims to provide an artificial intelligence micro-vibration detection method based on an FMCW millimeter wave radar, which aims to solve the problem that the traditional monitoring method can not realize accurate measurement of extremely tiny displacement change of a target. The embodiment of the invention is realized in such a way that an artificial intelligence micro-vibration detection method based on FMCW millimeter wave radar comprises the following steps: S1, adopting a self-adaptive optimization strategy to dynamically adjust the transmitting signal parameters of the FMCW radar and collect radar intermediate frequency signals; s2, preprocessing the acquired radar intermediate frequency signal, synchronously extracting the time-frequency domain physical characteristics and the deep learning abstract characteristics of the radar intermediate frequency signal, and carrying out cross-domain characteristic fusion to construct a fusion characteristic vector; S3, inputting the fusion feature vector into a light-weight transducer micro-vibration identification model trained by a domain self-adaptive framework for processing so as to obtain a micro-vibration detection and classification result; And S4, dynamically adjusting the reasoning precision and the calculation force configuration of the micro-vibration identification model according to the signal complexity indicated by the micro-vibration detection and classification result, and realizing the balance of the detection precision and the real-time performance. Preferably, in the step S1, the method includes the steps of: s11, calculating the signal-to-noise ratio of a received radar signal in real time, and taking the signal-to-noise ratio as an environmental state observation value of the reinforcement learning agent; s12, constructing a reinforcement learning model which takes the slope, bandwidth and period of a Chirp signal as an action space and the signal-to-noise ratio improvement and data effectiveness as a reward function; S13, outputting the current optimal Chirp signal modulation parameters through an interactive iterative optimization strategy network of the intelligent agent and the radar signal environment; s14, controlling the radar radio frequency front end according to the modulation parameters, and generating and transmitting FMCW signals adapting to the current environment and the target characteristics. Preferably, in the step S2, the method includes the st