A method and system for denoising a test vibration signal while drilling

Abstract

The present application belongs to the technical field of while-drilling signal processing optimization, and particularly discloses a while-drilling test vibration signal denoising method and system, which comprises the following steps: constructing an original signal with high amplitude and frequency at both ends and low amplitude and frequency in the middle according to the vibration characteristic difference between hard rock and soft rock or joints, and applying Gaussian noise with different signal-to-noise ratio levels to the original signal; processing the signal boundary by symmetric extension, decomposing the signal into low-frequency approximation coefficients and high-frequency detail coefficients of different scales by using a Daubechies filter, and introducing an adaptive threshold function to dynamically adjust the detail coefficients; iteratively searching in the global space by a genetic algorithm optimization engine to automatically match the optimal filter type, decomposition layer number and threshold parameter; applying the above parameters to the actual while-drilling test vibration signal to be processed, restoring the signal by inverse transformation to achieve the balance between signal fidelity and noise suppression. The present application effectively improves the fidelity of key features such as joints and cracks in a strong noise background.

Description

Technical Field

[0001] This invention belongs to the field of drilling signal processing and optimization technology, and more specifically, relates to a method and system for denoising vibration signals during drilling tests. Background Technology

[0002] Testing while drilling (WSD) is a crucial method for acquiring real-time rock mass information in geological exploration, playing a key role in tunneling, mine slope protection, and other engineering projects. Methods for inverting rock mass mechanics and structural characteristics based on drilling rig vibration signals have become a current research hotspot due to their advantages such as convenient data acquisition and high information density. However, engineering practice shows that test signals are susceptible to multiple interference sources, mainly including: mechanical vibration of the drilling rig itself, sensor coupling noise, and superposition of reflected waves from formation interfaces. The coupling effect of these multi-source interferences leads to severe ambiguity of signal characteristics, significantly reducing the reliability and engineering applicability of WWD data.

[0003] Traditional wavelet denoising methods are widely used in signal processing, but they still have significant limitations in practical engineering applications: key parameters such as threshold selection and the number of decomposition layers usually rely on empirical settings and lack adaptive adjustment mechanisms; fixed parameters are difficult to adapt to the complex and variable noise environment of drilling test data, easily leading to distortion of effective signals or noise residue; in particular, they are insufficient in capturing non-stationary signal features such as joints and fractures that suddenly appear during drilling. These problems seriously restrict the widespread application of this method in engineering practice.

[0004] Current research on signal fidelity control under multi-source noise coupling still has significant shortcomings: on the one hand, there is a lack of systematic quantitative evaluation frameworks; on the other hand, few studies organically combine intelligent algorithms such as genetic algorithms and Bayesian optimization with traditional denoising methods. To address these technical bottlenecks, an adaptive wavelet denoising method based on genetic algorithm parameter optimization is proposed. The optimal parameter combination is determined through multiple sets of simulated signal experiments, and then applied to the denoising processing of actual drilling test data. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method and system for denoising vibration signals during drilling tests. It combines the characteristics of multi-source interference during drilling tests, including mechanical vibration of the drilling rig itself, sensor coupling noise, and superposition of reflected waves from formation interfaces, with the high signal fidelity requirements of engineering projects such as tunnel excavation and mine slope protection. Accordingly, an adaptive processing method based on genetic algorithm parameter optimization is designed, constructing a complete closed-loop scheme from multi-condition signal simulation and time-frequency scale decomposition to global parameter optimization. In specific implementation, firstly, a reference signal reflecting different lithological characteristics is constructed through a simulation module, and multi-level noise is superimposed to cover complex working conditions; then, an adaptive transformation module is used to decompose the signal into components of different scales, and a dynamic adjustment mechanism is introduced to accurately suppress noise components; finally, an optimization engine iteratively searches within a multi-dimensional parameter space to automatically match the filter combination and processing intensity that best suits the current noise characteristics. Correspondingly, this solution can effectively solve the technical bottlenecks of traditional denoising methods that rely too much on human experience and have insufficient ability to capture non-stationary signals such as joints and fissures. At the same time, it also has the excellent effect of maintaining the integrity of the signal structure and high fidelity of features in the context of strong noise. Therefore, it is particularly suitable for high-precision applications such as advanced geological prediction of tunnel excavation, real-time monitoring of mine slopes and inversion of rock mechanics parameters.

[0006] To achieve the above objectives, according to one aspect of the present invention, a method for denoising vibration signals during drilling testing is proposed, comprising: Step 1: Based on the differences in vibration characteristics between hard rock and soft rock or joints during drilling tests, construct an original signal with high amplitude and frequency at both ends and low amplitude and frequency in the middle, and apply Gaussian noise at different signal-to-noise ratio levels to the original signal to obtain a multi-level simulated noisy signal. Step 2: The signal boundary is processed by symmetric extension. The signal is decomposed into low-frequency approximation coefficients and high-frequency detail coefficients at different scales using the Daubechies filter. An adaptive threshold function is introduced to dynamically adjust the detail coefficients. Step 3: With the objective function of minimizing the negative signal-to-noise ratio between the denoised signal and the original signal, global optimization is performed through initial population generation, fitness evaluation, selection, crossover and mutation operations to determine the optimal parameter combination for different wavelet basis functions, decomposition levels and threshold function parameters λ that control the shrinkage intensity. Step four: The optimal filter type, optimal decomposition level L, and optimal threshold parameter λ found by the genetic algorithm are applied to the actual drilling test vibration signal to be processed. The signal is recovered through inverse transformation to achieve a balance between signal fidelity and noise suppression.

[0007] As a further preferred embodiment, in step one, the original signal length is set to 20 seconds, and the acquisition frequency is 100Hz, expressed as: In the formula, x is the input parameter, referring to the original wavelet detail coefficients, and t is the time variable; The applied Gaussian noise signal-to-noise ratio levels include -5dB, 0dB, 5dB and 10dB, which correspond to four different levels of noise, to simulate the multi-source interference environment caused by the mechanical vibration of the drilling rig, sensor coupling noise and the superposition of reflected waves from the formation interface.

[0008] As a further preferred embodiment, in step two, based on the signal after symmetric extension processing, a Daubechies filter is used to decompose the signal into low-frequency approximation coefficients and high-frequency detail coefficients at different scales: In the formula, a j For low-frequency approximation coefficients, d j Let be the detail coefficients of the j-th layer, n be the index of the time-domain sampling point, and k be the discrete-time shift index.

[0009] As a further preferred embodiment, in step two, the adaptive threshold function dynamically adjusts the threshold strength by introducing an adjustment parameter λ. When the wavelet detail coefficients are close to the threshold, the function can retain more signal components to reduce the distortion of non-stationary signal features, especially for capturing joint and fracture signals that suddenly appear during drilling. The adaptive threshold function includes: In the formula, x is the input parameter, referring to the original wavelet detail coefficients, and y is the output parameter, referring to the detail coefficients after thresholding.

[0010] As a further preferred option, in step two, during the decomposition process, the low-frequency approximation coefficient α is maintained. j The high-frequency detail coefficients d of each layer remain unchanged. j The adaptive threshold function is applied for shrinkage processing. The processed coefficients are reconstructed using the inverse transform formula, and a segment with the same length as the original signal is truncated as the final denoised output.

[0011] As a further optimization, in step two, the detail coefficients d of the j-th layer are... j The expression for applying the adaptive threshold function to perform shrinkage processing is as follows: The reconstructed inverse transform formula is expressed as follows: The length of the original signal is truncated to obtain the denoised signal, and its performance is evaluated. The performance evaluation metrics include: In the formula, SNR, MSE, and Corr represent the signal-to-noise ratio, mean square error, correlation coefficient, and comprehensive index, respectively. These are the signal-to-noise ratio, mean square error, and correlation coefficient after standardization.

[0012] As a further preferred embodiment, in step three, the parameters of the genetic algorithm are configured as follows: the initial population size is set to 30, the maximum number of iterations is set to 150, and the crossover probability is set to 0.8; the search range of parameter λ is set between [0, 100]; the fitness function is directly related to the denoising performance evaluation index, that is, the population evolution is guided by maximizing the signal-to-noise ratio or minimizing the mean square error.

[0013] As a further preferred option, in step three, the selection of the optimal parameter combination also takes into account the correlation coefficient and the comprehensive index; wherein, the correlation coefficient is used to evaluate the effect of the denoised signal on the preservation of the original physical structure features, and the comprehensive index combines the signal fidelity and noise suppression effect to ensure that the optimal denoising gain can still be obtained by selecting a specific wavelet basis and a higher decomposition level in the context of strong noise.

[0014] As a further preferred embodiment, in step four, the value of the threshold parameter λ is negatively correlated with the noise intensity of the input signal; that is, when the signal to be processed exhibits strong noise, the genetic algorithm outputs a larger λ value to enhance the noise filtering capability; when the signal exhibits weak noise, it outputs a smaller λ value to preserve high-frequency details to the maximum extent.

[0015] According to another aspect of the present invention, a noise reduction system for vibration signals during drilling testing is also provided, comprising: Simulation module: used to construct the original signal simulating the vibration response while drilling, and to superimpose multi-level Gaussian noise to generate a training set; Adaptive Transform Module: Integrates wavelet decomposition, adaptive thresholding, and inverse transform functions for multi-scale signal processing; The genetic algorithm optimization engine is responsible for randomly generating a population and iteratively evolving it based on the signal-to-noise ratio (SNR) evaluation metric, searching for the optimal filter, decomposition level, and threshold parameters for the current signal characteristics; and Central processing unit: It is used to perform logical scheduling between the above modules, receive real-time drilling test data collected by sensors, and call the optimal parameters output by the optimization engine to perform denoising tasks, and finally output high-fidelity vibration signals for rock mass feature inversion.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages: 1. This invention achieves adaptive and intelligent optimization of denoising parameters, breaking through the bottleneck of experience-based reliance. Traditional wavelet denoising methods heavily depend on operator experience to set parameters, lacking universality. This invention, by introducing a genetic algorithm, establishes a complete "simulation-evaluation-optimization" closed-loop system, capable of automatically finding the optimal denoising parameters for different geological conditions, drilling equipment, and noise environments, completely eliminating reliance on manual experience and improving the automation level and engineering applicability of the method.

[0017] 2. This invention improves the fidelity of signal features under strong noise backgrounds. Because the optimization process of this invention directly aims to maximize the preservation of original signal features (i.e., similarity to the ideal simulated signal), it effectively balances the contradiction between noise suppression and signal preservation. Compared with traditional fixed-parameter methods, this invention has a stronger ability to capture and preserve non-stationary, weakly reflective signal features such as joints and fissures that suddenly appear during drilling, greatly reducing the distortion rate and noise residue of the effective signal, and providing a more reliable data foundation for subsequent inversion of rock mass mechanical parameters and identification of geological structures.

[0018] 3. This invention effectively addresses the complexity and variability of multi-source coupled noise in drilling testing. By proactively incorporating multi-source interference models, including mechanical vibration and sensor noise, into the simulation signal generation stage, this invention enables the optimization process to "anticipate" the complexity of the real-world environment. Therefore, the parameter combinations searched by the genetic algorithm inherently possess the ability to combat various coupled noise interferences, overcoming the shortcomings of traditional methods that rely on fixed parameters to adapt to complex and variable noise environments, and significantly improving the robustness and reliability of the denoising method.

[0019] 4. This invention provides key technical support for the engineering application of testing-while-drilling technology. It addresses the core pain points hindering the widespread adoption of this technology—data reliability and ease of processing. By providing a stable, efficient, and automated noise reduction solution, it lowers the technical threshold, enabling testing-while-drilling technology to more accurately and widely serve major projects such as advanced geological prediction for tunnel excavation and real-time monitoring of mine slope stability, effectively ensuring construction safety and efficiency. Attached Figure Description

[0020] Figure 1 This is a training flowchart of a method for denoising vibration signals during drilling, which is an embodiment of the present invention. Figure 2 This is a schematic diagram of the convergence result of the genetic algorithm involved in the embodiment of the present invention; Figure 3(a) in the figure is a schematic diagram of the denoising effect of different wavelet basis functions and decomposition levels for each noisy signal when SNR=-5; Figure 3 (b) in the diagram illustrates the denoising effect of different wavelet basis functions and decomposition levels on noisy signals when SNR=-0. Figure 3 (c) in the diagram illustrates the denoising effect of different wavelet basis functions and decomposition levels for each noisy signal when SNR=5. Figure 3 (d) in the figure is a schematic diagram of the denoising effect of different wavelet basis functions and decomposition levels for each noisy signal when SNR=10; Figure 4 It is a time-frequency domain comparison diagram of the original signal, noise signal and denoised signal at the same level. Figure 5 This is a time-frequency domain comparison diagram of the original signal, noise signal, and denoised signal at level 2. Figure 6 This is a time-frequency domain comparison diagram of the original signal, noise signal, and denoised signal at three horizontal time points; Figure 7 This is a time-frequency domain comparison diagram of the original signal, noise signal, and denoised signal at four different time periods. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] like Figure 1 As shown, the adaptive wavelet denoising algorithm provided in this embodiment of the invention mainly consists of four core steps: generating a simulated noisy signal, constructing an adaptive wavelet function, determining the optimization parameters of the genetic algorithm, and denoising using the optimal parameters. The specific implementation is as follows: Step 1: Generate a simulated noisy signal To simulate the physical differences between hard rock and soft rock / joints during actual drilling, this step first constructs a standard reference signal.

[0023] Original signal construction: Construct a discrete signal with a duration of 20 seconds and a sampling frequency of 100Hz. The signal is characterized by high amplitude and frequency at both ends (simulating hard rock) and low amplitude and frequency in the middle (simulating soft rock or joints).

[0024] Noise superposition: Gaussian white noise of different levels is applied to the original signal x[n] to generate analog signals with four signal-to-noise ratio (SNR) levels: Level 1 (-5dB), Level 2 (0dB), Level 3 (5dB) and Level 4 (10dB).

[0025] Step 2: Constructing the adaptive wavelet function. This step transforms the time-domain signal into the time-frequency domain using mathematical transformations and then performs nonlinear processing.

[0026] First, boundary processing and decomposition are performed: the signal length is extended to a power of 2 using the symmetric extension method to eliminate boundary effects. Then, multi-scale decomposition is performed using Daubechies filters (such as db5 and db6), calculated as follows: In the formula, n is the index of the original signal sampling point; k is the translation index after transformation; h and g are the low-pass and high-pass filters, respectively; a j d is the approximation coefficient. j This is the detail factor.

[0027] Threshold denoising: To balance noise suppression and signal fidelity, the detail coefficients d are... j Apply an adaptive threshold function. The expression for the adaptive threshold function is: x represents the original detail coefficients, y represents the processed coefficients, and λ represents the threshold parameter controlling the shrinkage intensity.

[0028] Signal reconstruction: preserving the approximation coefficient a j The signal remains unchanged, and the denoised signal is recovered through inverse transformation. In this step, the detail coefficients d of the j-th layer are... j The expression for applying the adaptive threshold function to perform shrinkage processing is as follows: The reconstructed inverse transform formula is expressed as follows: The length of the original signal is truncated to obtain the denoised signal, and its performance is evaluated. The performance evaluation metrics include: In the formula, SNR, MSE, and Corr represent the signal-to-noise ratio, mean square error, correlation coefficient, and comprehensive index, respectively. These are the signal-to-noise ratio, mean square error, and correlation coefficient after standardization.

[0029] Step 3: Optimize parameters using a genetic algorithm. In this step, a genetic algorithm (GA) is used to search for the optimal combination of denoising parameters in the global space. Details are as follows: (31) Population initialization: Randomly generate individuals containing wavelet basis type, decomposition level L and threshold parameter λ to form the initial population (size 30). (32) Fitness evaluation: The objective is to maximize the signal-to-noise ratio (SNR) of the denoised signal. The fitness function f(x) is defined as follows: In the formula, The original signal, This is the denoised signal.

[0030] (33) Evolutionary operation: Generate a new generation of individuals through selection, crossover (probability 0.8) and mutation operations until the maximum number of iterations of 150 is reached, and search for the parameter combination that makes the SNR reach the maximum value.

[0031] Step 4: Denoising using optimal parameters. The optimized parameters obtained from the optimization are then applied to the actual production data.

[0032] In this step, the optimal parameters are selected based on the estimated noise level of the actual signal. Generally, a larger λ value (e.g., 10.84) is needed in strong noise environments, along with a dB5 filter and 6-level decomposition. A smaller λ value (e.g., 0.61) is needed in weak noise environments to preserve weak signal characteristics.

[0033] Example 2 In this embodiment, as Figure 1 As shown, the denoising method for vibration signals during drilling testing includes four steps: generating a simulated noisy signal, constructing an adaptive wavelet function, determining the optimization parameters using a genetic algorithm, and denoising using the optimal parameters. Specifically: Step 1: Generate a noisy signal. In the vibration response during drilling testing, the amplitude and frequency of vibration in hard rock are significantly higher than those in soft rock or joints. Therefore, a raw signal with high amplitude and frequency at both ends and low amplitude and frequency in the middle is constructed. The signal duration is 20 seconds, and the acquisition frequency is 100Hz. The expression is: Gaussian noise with signal-to-noise ratios of -5, 0, 5, and 10 is applied to the original signal to obtain four levels of noisy signals (referred to as Level 1, Level 2, Level 3, and Level 4, respectively).

[0034] Step 2: Constructing the adaptive wavelet function. To avoid the boundary effects of wavelet transform, the signal length is extended to a power of 2 based on symmetric extension. A Daubechies filter is used to decompose the signal into low-frequency approximation coefficients and high-frequency detail coefficients at different scales. The calculation expression is as follows: In signal processing, to overcome the discontinuities near the threshold of hard thresholding functions and the signal attenuation problem of soft thresholding functions, an adaptive thresholding function is applied to dynamically adjust the threshold strength, better balance noise suppression and signal preservation, and retain more signal components and reduce signal distortion when the coefficients are close to the threshold. The calculation expressions are shown in Table 1, where x, λ, and y are the original wavelet detail coefficients, the threshold function controlling the shrinkage strength, and the detail coefficients after thresholding, respectively.

[0035] Table 1. Expressions for different threshold functions When using an adaptive threshold function, the threshold expression applied to the detail coefficients dj of the j-th layer is: Keeping the approximation coefficients aj unchanged, the signal is recovered by inverse transformation after updating the detail coefficients dj. The expression is: The length of the original signal is truncated to obtain the denoised signal, and its performance is evaluated. The performance indicators are shown in Table 2. In the formula... The original signal, For denoising signals, These are the signal-to-noise ratio, mean square error, and correlation coefficient after standardization.

[0036] Table 2 Noise Reduction Performance Evaluation Indicators Step 3: Optimize parameters using a genetic algorithm. The main steps are: randomly generate candidate solutions to form the initial population; use minimizing the negative signal-to-noise ratio as the objective function f(x) and perform fitness evaluation; select high-quality individuals for reproduction based on the evaluation results; simulate gene recombination to generate new individuals through crossover; randomly change some genes of individuals to increase population diversity; and find the optimal solution by controlling the number of iterations based on the maximum number of generations. The genetic algorithm parameters are shown in Table 3. The genetic algorithm training process for the level 3 signal using the db6 wavelet with a maximum decomposition level of 5 is as follows: Figure 2 As shown, the fitness function SNR has converged to its maximum value within the maximum number of iterations.

[0037] Table 3. Selection of Genetic Algorithm Parameters Step 4: Denoising using optimal parameters. For four levels of noisy analog signals, a genetic algorithm is used to optimize the value of λ based on different wavelet basis functions and different decomposition levels L, with the objective function being maximizing the signal-to-noise ratio. The denoising performance is then compared. Figure 3As shown in Table 3, the optimal filter, decomposition levels, and λ values ​​are obtained. A comparison of the time-frequency domain of the denoised signals obtained using the parameters in Table 3 is shown in the figure. Figure 3 As shown. By Figure 3 It is evident that the denoised signal obtained using the optimal parameters exhibits the highest signal-to-noise ratio (SNR), comprehensive index, and correlation coefficient, while also having a low mean square error. This indicates that the adaptive denoising algorithm effectively maintains signal fidelity while demonstrating good denoising performance and structure preservation capabilities. Table 4 shows that the SNR of the denoised signal is significantly better than that of the noisy signal. The value of parameter λ decreases as the SNR level increases; a larger λ value is suitable for processing strong noise signals, while a smaller λ value is suitable for processing weak noise signals. The parameters in Table 4 are named Level 1, Level 2, Level 3, and Level 4, from smallest to largest SNR. Figure 4 It is evident that the adaptive wavelet denoising algorithm has a good denoising effect on signals with various noise levels, and the time-frequency domain information of the denoised analog signal is closer to the original signal.

[0038] Table 4 Optimal Parameter Values ​​and Evaluation Indicators Example 3 Based on any of the above embodiments, or a combination of multiple embodiments, or a combination of multiple methods in the embodiments, this embodiment provides a drilling test vibration signal denoising system, including a simulation module, a feature extraction and transformation module, a genetic algorithm optimization engine, a signal reconstruction module, and a central processing unit. Each module communicates via a high-speed data bus to ensure real-time signal processing. Wherein: The simulation module is used to construct the original signal simulating the vibration response while drilling and to generate a training set by superimposing multi-level Gaussian noise. Specifically, the module constructs discrete signals based on the physical characteristics of high amplitude and frequency in hard rock and low amplitude and frequency in soft rock or joints during drilling tests. The signal duration was set to 20 seconds, and the acquisition frequency was 100Hz. To enable the system to resist multi-source interference, Gaussian noise was actively introduced into the original signal. By setting the signal-to-noise ratio (SNR) to -5, 0, 5, and 10, noisy training signals of four orders of magnitude, from "Level 1" to "Level 4," were generated. The noisy signal sequences generated by this module were used as input to the genetic algorithm optimization engine to simulate the superposition of mechanical vibration, sensor coupling noise, and ground interface reflected waves under real-world working conditions.

[0039] The adaptive transform module integrates wavelet decomposition, adaptive thresholding, and inverse transform functions for multi-scale signal processing. This module integrates wavelet decomposition operators and supports Daubechies filter banks. Specifically, this module includes: Wavelet decomposition units employ symmetrical extension techniques to handle signal boundaries, avoiding distortion caused by boundary effects. Convolution and downsampling operations are performed using Daubechies filter banks (such as db5 and db6). In the formula, n is the index of the original signal sampling point; k is the translation index after transformation; h and g are the low-pass and high-pass filters, respectively; a j d is the approximation coefficient. j This is the detail factor.

[0040] The dynamic threshold processing unit integrates an adaptive threshold function, overcoming the shortcomings of traditional hard thresholding (discontinuity) and soft thresholding (signal attenuation). Its function expression is: The module receives the threshold parameter λ determined by the optimization engine and processes the detail coefficients d for each layer. j Dynamic contraction is performed to filter out noise while accurately preserving non-stationary signal characteristics such as joints and fissures that suddenly appear during drilling.

[0041] Signal reconstruction unit, maintaining low-frequency approximation coefficient a j Without changing the original length, the signal is reconstructed using the inverse transform formula, and the original length is truncated to obtain a denoised high-fidelity vibration signal.

[0042] The genetic algorithm optimization engine is responsible for randomly generating a population and iteratively evolving it based on the signal-to-noise ratio (SNR) evaluation metric, searching for the optimal filter, decomposition level, and threshold parameters for the current signal characteristics. This module includes a population of 30 individuals, each containing three gene loci: filter type, decomposition level L, and threshold λ. Through 150 iterations of optimization, with SNR maximization as the feedback signal, the denoising strategy is dynamically adjusted. More specifically, this module executes: A population evolution mechanism is established. The engine sets the initial population size to 30 and generates candidate solutions through random encoding, including filter type (e.g., db5 or db6), decomposition level L (maximum 5 or 6), and threshold parameter λ (range [0, 100]). Fitness evaluation criteria are applied, with maximizing the signal-to-noise ratio (SNR) of the denoised signal as the core objective function. The engine calculates the denoised signal... With the original signal The SNR between the two values ​​guides the population to evolve towards higher fidelity. Convergence and iteration are performed, and population diversity is increased through gene recombination and mutation operations with a crossover probability of 0.8. Within a maximum of 150 iterations, the engine can converge the fitness function to its maximum value, thereby determining the optimal filter combination for the current noise level.

[0043] Central Processing Unit (CPU): This unit executes the logical scheduling between the aforementioned modules, receives real-time drilling test data collected by sensors, and calls upon the optimal parameters output by the optimization engine to perform denoising tasks. Ultimately, it outputs high-fidelity vibration signals for rock mass feature inversion. The CPU is responsible for managing the entire data lifecycle. The logic scheduling function is responsible for coordinating the simulation module to perform offline parameter pre-training and storing the trained optimal parameter table in the local cache; real-time data processing receives drilling vibration data collected by sensors, automatically identifies noise intensity, and calls the optimal parameter λ output by the optimization engine (such as λ=10.84 under strong noise and λ=0.61 under weak noise) to drive the adaptive transformation module.

[0044] Performance Quantitative Evaluation: The system calculates the mean square error (MSE) and correlation coefficient (r) in real time to ensure that the time-frequency domain information of the denoised signal is highly close to the original signal, providing a high-fidelity data foundation for the inversion of rock mechanics parameters in subsequent tunnel excavation, mine slope and other engineering projects.

[0045] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for denoising vibration signals during drilling tests, characterized in that, include: Step 1: Based on the differences in vibration characteristics between hard rock and soft rock or joints during drilling tests, construct an original signal with high amplitude and frequency at both ends and low amplitude and frequency in the middle, and apply Gaussian noise at different signal-to-noise ratio levels to the original signal to obtain a multi-level simulated noisy signal. Step 2: The signal boundary is processed by symmetric extension. The signal is decomposed into low-frequency approximation coefficients and high-frequency detail coefficients at different scales using the Daubechies filter. An adaptive threshold function is introduced to dynamically adjust the detail coefficients. Step 3: With the objective function of minimizing the negative signal-to-noise ratio between the denoised signal and the original signal, global optimization is performed through initial population generation, fitness evaluation, selection, crossover and mutation operations to determine the optimal parameter combination for different wavelet basis functions, decomposition levels and threshold function parameters λ that control the shrinkage intensity. Step four: The optimal filter type, optimal decomposition level L, and optimal threshold parameter λ found by the genetic algorithm are applied to the actual drilling test vibration signal to be processed. The signal is recovered through inverse transformation to achieve a balance between signal fidelity and noise suppression.

2. The method for denoising vibration signals during drilling testing according to claim 1, characterized in that, In step one, the original signal length is set to 20 seconds, and the sampling frequency is 100Hz, expressed as: ， In the formula, x is the input parameter, referring to the original wavelet detail coefficients, and t is the time variable; The applied Gaussian noise signal-to-noise ratio levels include -5dB, 0dB, 5dB and 10dB, which correspond to four different levels of noise, to simulate the multi-source interference environment caused by the mechanical vibration of the drilling rig, sensor coupling noise and the superposition of reflected waves from the formation interface.

3. The method for denoising vibration signals during drilling testing according to claim 1, characterized in that, In step two, based on the signal after symmetric extension processing, a Daubechies filter is used to decompose the signal into low-frequency approximation coefficients and high-frequency detail coefficients at different scales: ， ， In the formula, a j For low-frequency approximation coefficients, d j Let be the detail coefficients of the j-th layer, n be the index of the time-domain sampling point, and k be the discrete-time shift index.

4. The method for denoising vibration signals during drilling testing according to claim 1, characterized in that, In step two, the adaptive threshold function dynamically adjusts the threshold strength by introducing an adjustment parameter λ. When the wavelet detail coefficients are close to the threshold, the function can retain more signal components to reduce the distortion of non-stationary signal features, which is especially useful for capturing joint and fracture signals that suddenly appear during drilling. The adaptive threshold function includes: ， In the formula, x is the input parameter, referring to the original wavelet detail coefficients, and y is the output parameter, referring to the detail coefficients after thresholding.

5. The method for denoising vibration signals during drilling testing according to claim 1, characterized in that, In step two, during the decomposition process, the low-frequency approximation coefficient α is maintained. j The high-frequency detail coefficients d of each layer remain unchanged. j The adaptive threshold function is applied for shrinkage processing. The processed coefficients are reconstructed using the inverse transform formula, and a segment with the same length as the original signal is truncated as the final denoised output.

6. The method for denoising vibration signals during drilling testing according to claim 1, characterized in that, In step two, the detail coefficients d of the j-th layer are... j The expression for applying the adaptive threshold function to perform shrinkage processing is as follows: ， The reconstructed inverse transform formula is expressed as follows: ， The length of the original signal is truncated to obtain the denoised signal, and its performance is evaluated. The performance evaluation metrics include: ， ， ， ， In the formula, SNR, MSE, and Corr represent the signal-to-noise ratio, mean square error, correlation coefficient, and comprehensive index, respectively. These are the signal-to-noise ratio, mean square error, and correlation coefficient after standardization.

7. The method for denoising vibration signals during drilling testing according to claim 1, characterized in that, In step three, the parameters of the genetic algorithm are configured as follows: the initial population size is set to 30, the maximum number of iterations is set to 150, and the crossover probability is set to 0.8; the search range of parameter λ is set between [0, 100]; the fitness function is directly related to the denoising performance evaluation index, that is, the population evolution is guided by maximizing the signal-to-noise ratio or minimizing the mean square error.

8. The method for denoising vibration signals during drilling testing according to claim 1, characterized in that, In step three, the selection of the optimal parameter combination also takes into account the correlation coefficient and comprehensive index; Among them, the correlation coefficient is used to evaluate the effect of the denoised signal on the preservation of the original physical structure features, while the comprehensive index combines signal fidelity and noise suppression effect to ensure that the optimal denoising gain can still be obtained by selecting specific wavelet bases and higher decomposition levels in the context of strong noise.

9. The method for denoising vibration signals during drilling testing according to claim 1, characterized in that, In step four, the value of the threshold parameter λ is negatively correlated with the noise intensity of the input signal; that is, when the signal to be processed is characterized by strong noise, the genetic algorithm outputs a larger λ value to enhance the noise filtering capability; when the signal is characterized by weak noise, it outputs a smaller λ value to preserve high-frequency details to the maximum extent.

10. A noise reduction system for vibration signals during drilling testing, characterized in that, include: Simulation module: used to construct the original signal simulating the vibration response while drilling, and to superimpose multi-level Gaussian noise to generate a training set; Adaptive Transform Module: Integrates wavelet decomposition, adaptive thresholding, and inverse transform functions for multi-scale signal processing; The genetic algorithm optimization engine is responsible for randomly generating a population and iteratively evolving it based on the signal-to-noise ratio (SNR) evaluation metric, searching for the optimal filter, decomposition level, and threshold parameters for the current signal characteristics; and Central processing unit: It is used to perform logical scheduling between the above modules, receive real-time drilling test data collected by sensors, and call the optimal parameters output by the optimization engine to perform denoising tasks, and finally output high-fidelity vibration signals for rock mass feature inversion.

Images