A PDC bit residual life prediction method based on torsional time series data

Abstract

The present application relates to the technical field of drilling operation, and particularly relates to a PDC bit residual life prediction method based on torsion time series data, which comprises the following steps: collecting torsion data of the bit to construct torsion time series data; extracting wear modal components by using a variational mode decomposition algorithm containing a frequency band bandwidth constraint term and a time domain sparsity constraint term; calculating the logarithmic energy features and multi-scale spectrum entropy of the wear modal components and weighting fusion to obtain a composite degradation index; calculating the ratio of the composite degradation index to the cumulative running time, multiplying by a preset wear acceleration factor to obtain an instantaneous degradation rate, and combining with a preset failure threshold to obtain the residual life. The present application effectively extracts the wear modal components by introducing the time domain sparsity constraint term, and constructs a nonlinear prediction model in combination with the wear acceleration factor, thereby improving the accuracy of the PDC bit residual life prediction.

Description

A method for predicting the remaining life of PDC drill bits based on torque time series data Technical Field

[0001] This invention relates to the field of drilling operation technology, and in particular to a PDC drill bit remaining life prediction method based on torque time series data. Background Technology

[0002] Polycrystalline diamond composite (PDC) drill bits are core rock-breaking tools in oil and gas drilling operations. Their cutting teeth continuously interact with the formation rock, and their working condition directly affects the drilling speed and well construction costs. Failure to accurately and promptly monitor their health can lead to decreased drilling efficiency and even engineering accidents such as broken teeth falling out downhole, resulting in economic losses. Therefore, monitoring the condition and assessing the remaining life of PDC drill bits based on real-time data generated during drilling has significant application value.

[0003] Existing life prediction techniques typically rely on laboratory bench tests or field logging data, and empirical mode decomposition (EMD) algorithms are often used to process torque time-series data in the signal processing stage. These methods primarily construct upper and lower envelopes based on the distribution of local extrema of the signal in the time domain. Through iterative sieving, the complex raw torque signal is decomposed into several intrinsic mode function components with different characteristic scales, thereby separating high-frequency impact characteristics that reflect the wear state of the drill bit. These characteristics are then used to construct degradation indices for trend prediction.

[0004] However, commonly used adaptive mode decomposition algorithms have limitations when processing noisy torque signals: due to the random rock fracturing impact and mechanical rotation background vibration during downhole drilling, the torque signal is mixed with a large amount of irregular noise, causing unstable drift in the distribution of local extrema. This uncertainty causes envelope fitting distortion, which in turn leads to mode aliasing problems. This results in weak high-frequency impact components representing early wear characteristics being incorrectly classified into low-frequency components, or the same component being mixed with fluctuations at different time scales. This incomplete signal decomposition causes the characteristic components to lose their single physical meaning, making it difficult to truly reflect the progressive wear process of the drill bit and affecting the accuracy of remaining life prediction. Summary of the Invention

[0005] To address the technical problem of existing adaptive mode decomposition algorithms suffering from mode aliasing due to extreme point drift under strong noise interference, resulting in distorted wear feature extraction and difficulty in accurately representing the true wear state of drill bits, this invention provides a PDC drill bit remaining life prediction method based on torque time series data. This method includes the following steps:

[0006] Torque data of the drill bit is collected at a preset sampling frequency to construct torque time-series data. Based on the torque time-series data, wear mode components are extracted using a variational mode decomposition algorithm. The variational mode decomposition algorithm is based on a variational objective functional, which includes constraint terms for constraining the bandwidth of the mode components and constraint terms for enhancing the temporal sparsity of the mode components. Taking any moment as the current moment, the logarithmic energy feature and multi-scale spectral entropy of the wear mode components are calculated. The logarithmic energy feature and multi-scale spectral entropy are weighted and fused to obtain the composite degradation index at the current moment. The ratio of the composite degradation index to the cumulative running time corresponding to the current moment is calculated. The ratio is multiplied by a preset wear acceleration factor to obtain the instantaneous degradation rate. The difference between the composite degradation index and a preset failure threshold is obtained. The difference is divided by the instantaneous degradation rate to obtain the remaining life of the PDC drill bit.

[0007] This invention introduces bandwidth constraints and temporal sparsity constraints into variational mode decomposition, leveraging the difference between the sparsity of wear impact and the continuity of background vibration to effectively separate wear characteristics and reduce mode aliasing. Simultaneously, this invention integrates logarithmic energy and multi-scale spectral entropy features to construct a composite degradation index from both impact intensity and signal complexity dimensions. Furthermore, the instantaneous degradation rate is calculated using the ratio of this index to cumulative operating time and the wear acceleration factor, achieving nonlinear lifetime prediction. This method conforms to the self-accelerating law of drill bit wear and, compared to traditional linear models, better reflects the accelerated degradation trend in later stages, providing a more conservative and safe lifetime assessment in the field and reducing the risk of downhole accidents.

[0008] Preferably, the variational objective functional satisfies the following relation:

[0009] ;

[0010] in, Based on full set of modal components With the central frequency of the entire set The constructed variational objective functional to be minimized; It is the preset number of modal component decomposition layers; It is the sampling time; It is the first The modal components to be solved at the sampling time The amplitude; It is the first The center frequencies corresponding to the modal components to be solved; It is a balance parameter; It is the L2 norm; It is an L1 norm; Indicates the sampling time Find the partial derivative; It is the Dirac function; It is a temporal convolution operator; It is the imaginary unit.

[0011] This invention constructs a variational objective functional that includes bandwidth and sparsity constraints, mathematically constraining both the frequency and time domain morphologies of modal components. Specifically, the L2 norm of the gradient constrains the bandwidth of the modal components, keeping them compact in the frequency domain to represent a single vibration mode. The L1 norm is used as a sparsity constraint to induce sparsity in the time domain. This dual-constraint mechanism leverages the fundamental morphological differences between wear impact signals and mechanical background noise, forcing the optimization process to retain the sparse wear signal in specific components while distributing the continuous, smooth background noise to other components. This enables accurate separation and extraction of subtle wear features even under strong noise interference.

[0012] Preferably, solving the variational objective functional includes: converting the variational objective functional into an unconstrained optimization problem using the alternating direction multiplier method, performing iterative calculations in the frequency domain, and filtering several modal components from the iterative calculations to obtain the wear modal components.

[0013] Preferably, the step of filtering the iteratively calculated modal components to obtain the wear modal components includes: calculating the kurtosis values ​​of the several modal components and selecting the modal component with the largest kurtosis value as the target wear modal component. The calculation of kurtosis values ​​is existing technology and will not be elaborated upon here.

[0014] Preferably, the logarithmic energy characteristic satisfies the following relation:

[0015] ;

[0016] in, The wear mode component at the sampling time Logarithmic energy characteristics; The wear mode component at the sampling time The amplitude; It is the length of the integration window; It is the initial reference energy; It is an integral variable; It is a logarithmic function.

[0017] This invention obtains logarithmic energy characteristics by calculating the cumulative energy value of the wear modal components within the integration window and performing a logarithmic transformation. It uses local integration operations to smooth the random fluctuation noise of single-point signals, preserves the energy accumulation characteristics of impact signals, and uses a logarithmic function to compress the dynamic range of energy values, reducing the impact of dimensional differences. This allows the feature to sensitively reflect the increase in impact intensity of the drill bit relative to the initial reference energy in a dimensionless form, improving the index's ability to identify early minor wear.

[0018] Preferably, the acquisition of the multi-scale spectral entropy includes: performing probability statistics on the wear mode components to obtain their energy probability distribution at several preset frequency scales, calculating the entropy value of the energy probability distribution based on Shannon information entropy, and obtaining the multi-scale spectral entropy.

[0019] Preferably, the step of weightedly fusing the logarithmic energy feature with the multi-scale spectral entropy to obtain the composite degradation index at the current moment includes: calculating the coefficient of variation of the logarithmic energy feature within a preset time window to obtain a first coefficient of variation; calculating the coefficient of variation of the multi-scale spectral entropy within the preset time window to obtain a second coefficient of variation; normalizing the first coefficient of variation and the second coefficient of variation using the sum of the first coefficient of variation and the second coefficient of variation to obtain the logarithmic energy feature weight and the multi-scale spectral entropy weight; and weightedly summing the logarithmic energy feature and the multi-scale spectral entropy feature using the logarithmic energy feature weight and the multi-scale spectral entropy weight to obtain the composite degradation index.

[0020] This invention utilizes the logarithmic energy characteristics and the coefficient of variation of multi-scale spectral entropy within a preset time window to calculate weights, establishing an objective data-driven weighting mechanism. This mechanism is based on the statistical characteristic that the coefficient of variation reflects the degree of data dispersion, and can automatically assess the degree of fluctuation of each feature at the current stage. When a feature fluctuates significantly due to a fault, its corresponding coefficient of variation and weight will automatically increase, thereby ensuring that the composite degradation index can dynamically focus on the most sensitive fault feature at the current moment, taking into account the signal change patterns at different stages throughout the drill bit's life cycle.

[0021] Preferably, the remaining lifetime satisfies the following relationship:

[0022] ;

[0023] in, It is the time when the drill bit is sampling The remaining lifespan; It is a preset failure threshold; Sampling time The composite degradation index; From the start point of data acquisition to the sampling time The cumulative time; It is a wear-accelerating factor; This is a preset correction option.

[0024] This invention constructs a nonlinear remaining life prediction formula with an embedded wear acceleration factor. The average degradation rate is obtained by calculating the ratio of the composite degradation index to the cumulative operating time, and the wear acceleration factor is used to correct the average rate to obtain the instantaneous degradation rate. This model mathematically simulates the self-accelerating dynamic characteristics of the wear process. It uses the instantaneous degradation rate instead of the constant average rate to estimate the remaining life, which reduces the error of the traditional linear model overestimating the remaining life in the later stage of wear, and makes the prediction results more consistent with the actual physical law of nonlinear acceleration of drill bit wear over time.

[0025] Preferably, the step of collecting torque data of the drill bit at a preset sampling frequency to construct torque time-series data includes: collecting torque signals using a high-frequency torque sensor; setting a sampling frequency, and using a low-pass filter to remove high-frequency electromagnetic noise greater than half the sampling frequency from the torque signal according to the Nyquist sampling theorem; extracting stable drilling stage data of a preset duration and arranging them according to the order of sampling times to obtain torque time-series data.

[0026] Preferably, the method further includes a graded early warning system, which includes: setting a first early warning threshold and a second early warning threshold, wherein the first early warning threshold is less than the second early warning threshold and both are less than a preset failure threshold; when the composite degradation index is greater than the first early warning threshold, a mild wear warning signal is generated; when the composite degradation index is greater than the second early warning threshold, a severe wear alarm signal is generated and it is recommended to replace the drill bit.

[0027] The beneficial effects of this invention are as follows: This invention introduces constraint terms to the variational objective functional of the variational mode decomposition algorithm to constrain the bandwidth of the modal components and to enhance the temporal sparsity of the modal components. Utilizing the physical difference between the time-domain sparsity of the transient impact signal generated by PDC drill bit wear and the continuity of the background vibration signal, the wear features are separated from complex background noise. This reduces the impact of mode aliasing on feature extraction and ensures the physical purity of the extracted wear modal components. Subsequently, this invention calculates the logarithmic energy characteristics and multi-scale spectral entropy of the wear modal components and performs weighted fusion to obtain the wear modal components from the impact strength... This invention comprehensively evaluates the wear state of drill bits from two dimensions: degree and signal complexity, and obtains a composite degradation index. Furthermore, this invention uses the ratio of the composite degradation index to the cumulative running time corresponding to the current moment, combined with a preset wear acceleration factor, to obtain the instantaneous degradation rate. This nonlinear extrapolation method conforms to the self-accelerating thermodynamic law of PDC drill bit wear rate increasing with time. Compared with the traditional linear extrapolation model, this invention can better fit the accelerated degradation trend of drill bits in the later stage of wear, thereby providing field operators with a more conservative remaining life assessment result that conforms to actual working conditions, reducing the risk of downhole tooth breakage or drill bit loss due to prediction deviation. Attached Figure Description

[0028] Figure 1 is a flowchart of a PDC drill bit remaining life prediction method based on torque time series data provided by an embodiment of the present invention;

[0029] Figure 2 is a schematic diagram of drill bit torque timing data provided in an embodiment of the present invention;

[0030] Figure 3 is a schematic diagram of the wear mode components of the drill bit provided in an embodiment of the present invention;

[0031] Figure 4 is a schematic diagram of the correlation analysis between the composite degradation index and the predicted remaining life of the drill bit provided in the embodiment of the present invention. Detailed Implementation

[0032] This invention provides a method for predicting the remaining life of a PDC drill bit based on torque time-series data, as shown in Figure 1. The method includes steps S100-S400:

[0033] Step S100: Collect torque data of the drill bit at a preset sampling frequency to construct torque time series data.

[0034] It should be noted that during downhole drilling, the torque signal of a PDC drill bit is often mixed with high-frequency electromagnetic noise generated by wellbore friction, drill string collision, and electrical equipment. The frequencies of these noises are typically much higher than the mechanical vibration frequencies generated by the drill bit cutting the rock. Direct analysis without processing will severely interfere with the accuracy of subsequent feature extraction, leading to misjudgments of wear conditions. Therefore, this invention first purifies the raw signal before analysis and discretizes it into a standard time series to ensure that the signal-to-noise ratio and temporal structure of the input data meet the analysis requirements.

[0035] Specifically, firstly, a high-frequency torque sensor is installed on a torsion impact test platform or on-site drilling rig. The PDC drill bit to be tested is mounted on the test bench, and simulated drilling parameters, including rotational speed, drilling pressure, and rock sample strength, are set. The test platform is then started, and the raw torque signal during the drilling process is acquired in real time. Next, the sampling frequency is set, preferably between 1kHz and 5kHz, to cover the frequency band of the high-frequency impact characteristics of the PDC drill bit's cutting teeth. Based on the Nyquist sampling theorem, a low-pass filter is used to remove high-frequency electromagnetic noise above half the system sampling frequency from the raw signal, and data from a stable drilling phase of a preset duration is extracted as the analysis sample. Finally, the extracted discrete data points are arranged according to the chronological order of sampling time to obtain the torque time-series data.

[0036] Figure 2 shows a schematic diagram of drill bit torque time series data. The horizontal axis represents time, and the vertical axis represents torque magnitude. The curve in the figure shows the torque change trend of the drill bit throughout its entire life cycle. The initial waveform shows regular periodic fluctuations with stable amplitude, corresponding to the stable operation stage. The later waveform is superimposed with irregular spikes, and the oscillation amplitude is significantly increased, which intuitively reflects the abnormal vibration characteristics of the drill bit caused by wear accumulation.

[0037] At this point, the torque timing data has been obtained.

[0038] Step S200: Based on torque time series data, extract wear mode components using variational mode decomposition algorithm; the variational mode decomposition algorithm is based on variational objective functional, which includes constraint terms for constraining the bandwidth of the mode components and constraint terms for enhancing the temporal sparsity of the mode components.

[0039] It should be noted that the localized breakage or wear of the cutting teeth of the PDC drill bit manifests physically as a series of transient impact pulses with a specific center frequency. These signals exhibit significant sparsity in the time domain, meaning their amplitude is low for most of the time, with a sudden amplitude change only occurring at the moment of damage. In contrast, the background signal of the drill rig's mechanical rotation exhibits continuous quasi-periodic fluctuations, appearing as a narrowband signal in the frequency domain. To separate these two distinct signal types, this invention employs the framework of Variational Mode Decomposition (VMD). VMD is an adaptive, non-recursive signal processing technique whose core advantage lies in its ability to transform complex signal decomposition problems into variational constraint problems, thereby effectively reducing the mode aliasing phenomenon commonly found in traditional empirical mode decomposition. Based on this, this invention utilizes the core idea of ​​the VMD framework, namely, presumably, the original signal is composed of... The signal is composed of multiple amplitude-modulated and frequency-modulated sub-signals with different center frequencies, i.e., modal components superimposed. The number of decomposition layers of the modal components needs to be set according to the complexity of the signal, preferably in the range of 3 to 8. In this embodiment, it is preferably set to 5 to effectively separate background vibration, wear impact, and random noise. By constructing a mathematical optimization model, the decomposition layer is solved in reverse. The unknown modal components and their corresponding center frequencies are used to accurately extract specific wear characteristics.

[0040] Specifically, for the torque time series data, the present invention transforms its decomposition problem into a constrained optimization problem, and obtains the wear mode components by minimizing the variational objective functional.

[0041] First, we define bandwidth and sparsity constraints. It should be noted that, considering that the mechanical vibration modes of drill bits are usually narrowband signals in the frequency domain, in order to extract vibration features with clear physical meaning, the constructed model should constrain the decomposed components to remain compact in the frequency domain to avoid spectral dispersion. At the same time, considering that the early wear of cutting teeth is a transient impact in the time domain, which has a significant morphological difference from the continuous background vibration, the constructed model should also constrain the decomposed components to remain sparse in the time domain to accurately capture weak impact pulse features.

[0042] Based on the above logic, the variational objective functional satisfies the following relation:

[0043] ;

[0044] in, Based on full set of modal components With the central frequency of the entire set The constructed variational objective functional to be minimized; It is the preset number of modal component decomposition layers; It is the sampling time; It is the first The modal components to be solved at the sampling time The amplitude; It is the first The center frequencies corresponding to the modal components to be solved; It is a balance parameter; It is the L2 norm; It is an L1 norm; Indicates the sampling time Find the partial derivative; It is the Dirac function; It is a temporal convolution operator; It is the imaginary unit, and its value is .

[0045] In this relation, This is a bandwidth constraint term, which utilizes the Hilbert transform operator. With modal components Perform convolution operations to construct an analytic signal with a single-sided spectrum to eliminate interference from negative frequency components; multiply by an exponential term. The spectral center of the analyzed signal is moved from... The signal is then moved to the baseband zero frequency. Finally, the smoothness of the baseband signal is calculated using differentiation combined with the L2 norm. Since a smaller gradient energy of the baseband signal means a tighter distribution of its original signal around the center frequency, this term achieves the constraint of minimizing the bandwidth of the modal components. As the sparsity constraint term, due to the mathematical property of the L1 norm inducing sparse solutions, this term makes the decomposed modal components... As sparse as possible in the time domain. When properly configured, the minimization process of the variational objective functional tends to preserve the wear signal with sparse impact characteristics within a specific range. In this process, continuous background noise is distributed to other components, thereby achieving effective separation between the two.

[0046] It should be added that the balance parameters The value needs to be set according to the signal-to-noise ratio of the signal. When the background noise is strong, it can be increased appropriately. Values, for example, ranging from 1000 to 2000, are used to enhance the suppression of background noise and highlight sparse impact components; when the signal is relatively pure, the value can be reduced. The value, for example, is between 500 and 1000, to retain more detailed information. In this embodiment, The preferred setting is 1500.

[0047] Then, in order to obtain the optimal solution, the present invention uses the Alternating Direction Multiplier Method (ADMM) to perform iterative solution in the frequency domain.

[0048] Specifically, since the variational objective functional contains convolution operations, the computational cost of directly solving it in the time domain is large. Therefore, this invention utilizes Passevar's theorem and the properties of Fourier transform to transfer the solution process to the frequency domain.

[0049] During the solution process, the modal components, center frequency, and Lagrange multipliers to be solved are first initialized, followed by an alternating iterative loop. In each iteration, the multivariate optimization problem is decomposed into the following subproblems and solved sequentially: the center frequency and Lagrange multipliers are treated as known constants, and the modal components are updated in the frequency domain to simultaneously satisfy bandwidth minimization and sparsity constraints; based on the updated modal components, the center frequency is updated by calculating the centroid of the current modal power spectrum to ensure the compactness of the spectrum; the multipliers are updated according to the residuals to ensure the convergence of the constraints. This alternating update process continues until the objective function value or the change in modal components between two adjacent iterations meets a preset convergence threshold. At this point, the optimal solution in the frequency domain is output, and it is converted into the time domain by inverse Fourier transform. One modal component.

[0050] Finally, the wear mode components are screened. After obtaining the decomposed components... After identifying the modal components, it is necessary to filter out the modal components that contain the main wear information. Considering that wear and micro-chipping of PDC drill bits are usually accompanied by non-stationary impact signals, and kurtosis is an effective statistical indicator for characterizing the impact characteristics of the signal, it is highly sensitive to deviations from the normal distribution.

[0051] Specifically, calculate the following respectively. The kurtosis values ​​of each modal component are used, and the modal component with the largest kurtosis value is selected as the wear modal component, denoted as . .

[0052] Figure 3 shows a schematic diagram of the wear modal components of the drill bit. The horizontal axis represents time, and the vertical axis represents the amplitude of the wear component. The curves in the figure show the extracted modal components: the values ​​of the first half of the curve remain near zero, indicating no obvious wear; the second half of the curve shows pulse-like abrupt changes with increasing frequency and intensity.

[0053] At this point, the wear mode components at each moment have been obtained.

[0054] Step S300: Taking any time as the current time, calculate the logarithmic energy characteristics and multi-scale spectral entropy of the wear mode components, and perform weighted fusion of the logarithmic energy characteristics and multi-scale spectral entropy to obtain the composite degradation index at the current time.

[0055] It should be noted that a single signal feature is insufficient to cover the wear state of a PDC drill bit throughout its entire life cycle. For example, in the early stages of tooth breakage, the signal mainly manifests as violent fluctuations in transient impact energy; while in the later stages of wear, as the wear surface expands, the frequency distribution of the signal becomes more chaotic, and the change in complexity becomes more pronounced. To comprehensively evaluate the drill bit condition, this invention first obtains the energy index characterizing impact intensity and the entropy index characterizing complexity, and then fuses them using an objective weighting mechanism.

[0056] First, when calculating the logarithmic energy characteristics, it's important to note that early damage to PDC drill bits, such as micro-chipping, physically manifests as high-frequency transient energy release. Using only the squared amplitude of a single sampling point as the energy indicator is susceptible to random noise and exhibits excessive fluctuations, failing to reflect the true impact intensity. Therefore, this invention employs a short-time energy flow construction logic, smoothing random noise through local integration in the time dimension while preserving the energy accumulation characteristics of high-frequency impacts, thereby obtaining a more stable and physically representative instantaneous energy metric.

[0057] Specifically, taking any given moment as the current moment and the current moment as the cutoff point, an integral window is constructed by backtracking backward. , The length of the integration window is defined. The cumulative energy value of the wear mode components within the neighborhood window at the current time is calculated, compared with the initial reference energy, and subjected to a logarithmic transformation to obtain the logarithmic energy characteristics at any given time. The length of the integration window is defined as follows. It needs to be able to cover the duration of a single impact pulse to ensure the integrity of energy capture; in this embodiment, it is preferably set to 5 to 20.

[0058] Based on the above logic, the wear mode components at the sampling time logarithmic energy characteristics Satisfying the relation:

[0059] ;

[0060] in, The wear mode component at the sampling time The amplitude; It is the length of the integration window; It is the initial reference energy, preferably taken during the initial break-in stage of the drill bit entering the well, such as the average energy value in the first 10 minutes. It is an integral variable; It is a logarithmic function.

[0061] In this relationship, the molecule Calculate sampling time The instantaneous energy of the wear component reflects the intensity of the impact; the denominator The absolute dimension difference of energy is eliminated; the logarithmic operation plays a role in numerical smoothing and dynamic range compression, so that the logarithmic energy characteristics can reflect the impact energy increment of the drill bit relative to the initial state in a dimensionless form.

[0062] Then, multi-scale spectral entropy eigenvalues ​​are obtained. It should be noted that as wear intensifies, the interaction surface between the drill bit and the rock changes from point contact to surface contact, leading to an increase in the nonlinearity of the frictional vibration signal and a shift in the spectral distribution from a singular to a chaotic and disordered state. Simple energy indicators are insufficient to fully characterize this complex change in frequency domain structure. Therefore, this invention introduces information entropy theory to assess the degree of signal disorder as a complementary dimension for evaluating the wear state.

[0063] Specifically, probabilistic statistics are performed on the wear mode components to obtain their energy probability distributions at different frequency scales or amplitude ranges; the entropy value of this probability distribution sequence is calculated based on the Shannon information entropy principle, and this is used as the multi-scale spectral entropy. The larger this value, the more uniform and chaotic the energy distribution of the signal, corresponding to a more severe physical state of drill bit wear. The calculation of Shannon information entropy is existing technology and will not be elaborated upon here.

[0064] Finally, a composite degradation index is constructed based on adaptive variation weighting. It should be noted that logarithmic energy features and multi-scale spectral entropy exhibit different sensitivities at different wear stages. Using fixed weights for fusion cannot adequately account for feature changes throughout the entire cycle. Therefore, this invention introduces a weighting mechanism based on the coefficient of variation. This mechanism leverages the statistical property of the coefficient of variation, which reflects the degree of data dispersion, to assess the importance of each feature at the current moment, thereby constructing a unified composite degradation index.

[0065] Specifically, firstly, a long-term neighborhood window is set. The coefficient of variation of the logarithmic energy feature within a preset long-term neighborhood window is calculated to obtain the first coefficient of variation; the coefficient of variation of the multi-scale spectral entropy within the long-term neighborhood window is calculated to obtain the second coefficient of variation. It should be noted that, unlike the integral window, the long-term neighborhood window here is used to capture the statistical fluctuation patterns of the feature sequence on a macroscopic time scale. Sliding time window The size needs to be set according to the real-time requirements of the drilling operation: for scenarios with complex geological conditions and rapid changes in drill bit status, it can be appropriately reduced. Values, such as the number of data points in 30 seconds, can be set to improve the response speed to sudden anomalies; for scenarios with homogeneous formations and a stable drilling process, the value can be appropriately increased. Values, such as the number of data points over 60 seconds, are used to reduce weight jitter caused by random interference. In this embodiment, considering both sensitivity and stability, it is preferable to use... Set the number of data points to 45 seconds.

[0066] Then, using the sum of the first and second coefficients of variation, the first and second coefficients of variation are normalized to obtain the logarithmic energy feature weight and the multi-scale spectral entropy weight; the logarithmic energy feature and the multi-scale spectral entropy feature are then weighted and summed using the logarithmic energy feature weight and the multi-scale spectral entropy weight to obtain the composite degradation index. It should be noted that the dynamic weighting mechanism based on the proportion of the coefficient of variation ensures that the composite index can focus on the most sensitive fault feature at the current moment: when a certain feature, such as the early impact energy, fluctuates drastically, its coefficient of variation increases, thus taking a dominant position in the fusion result and achieving sensitive capture of abnormal drill bit conditions.

[0067] Thus, the composite degradation index was obtained.

[0068] Step S400: Calculate the ratio of the composite degradation index to the cumulative running time corresponding to the current moment, multiply the ratio by a preset wear acceleration factor to obtain the instantaneous degradation rate; obtain the difference between the composite degradation index and the preset failure threshold, divide the difference by the instantaneous degradation rate to obtain the remaining life of the PDC drill bit.

[0069] It should be noted that the wear process of PDC drill bits exhibits a significant self-accelerating characteristic. That is, as the wear plateau area expands, frictional heat accumulation leads to a decline in the material properties of the cutting teeth, causing the wear rate to increase nonlinearly in a power-law manner. Existing linear extrapolation models typically assume a constant wear rate and are suitable for the stable wear stage. However, when dealing with the later stages of wear characterized by acceleration, their accuracy in predicting remaining life still has room for improvement. Therefore, this invention provides a nonlinear prediction model with an embedded wear acceleration factor. This model captures the dynamic law of wear rate increasing over time and directly utilizes the instantaneous rate of change relationship of the composite degradation index for extrapolation.

[0070] First, the average degradation rate at the sampling time is calculated. It should be noted that in order to predict future degradation trends, it is first necessary to assess the overall wear rate of the drill bit from its entry into the well to the present. Although the average rate cannot directly represent the current acceleration state, it contains the basic wear efficiency of the drill bit under the current formation hardness and is the basic benchmark for constructing the instantaneous rate estimation model.

[0071] Specifically, the cumulative time from the start point of data collection to the current moment is obtained, and the average degradation rate at the current moment is obtained by dividing the composite degradation index value at the sampling moment by the cumulative time.

[0072] Then, a remaining life prediction model with an embedded acceleration factor is constructed. It should be noted that, considering the self-accelerating nature of wear, the instantaneous wear rate of the drill bit at the current moment is usually higher than the average degradation rate mentioned above. To eliminate the error of linear extrapolation, a correction factor conforming to the physical wear law needs to be introduced to map the average rate to the instantaneous rate. Based on the power-law degradation model, there is a multiple relationship between the instantaneous rate and the average rate; this multiple is the wear acceleration factor. Using the corrected instantaneous rate as the denominator and the remaining degradation margin as the numerator, the remaining life that conforms to actual working conditions is obtained.

[0073] Based on the above logic, the drill bit at the sampling time Remaining lifespan Satisfying the relation:

[0074] ;

[0075] in, It is a preset failure threshold; Sampling time The composite degradation index; It is the cumulative time from the start point of data collection to the current moment; the remaining lifetime prediction calculation is only performed within... Performed when >0; It is a wear-accelerating factor; It is a preset correction term, the dimensions of which are the same as the average degradation rate. The dimensions are consistent, which is used to avoid the denominator being zero and to ensure numerical stability.

[0076] In this relationship, the molecule Represents the remaining degradation margin from failure; denominator Using the average degradation rate at the current moment The product of this factor and the acceleration factor provides an estimate of the current instantaneous wear rate. The physical derivation of this relationship originates from the power-law degradation model. The derivative property of the instantaneous rate is expressed as This eliminates the unknown model parameters. Therefore, the prediction model avoids dependence on fitting to historical data throughout the entire cycle and can dynamically extrapolate future nonlinear accelerated wear trajectories based on the current degradation state and cumulative time.

[0077] It should be added that the failure threshold... This value is derived from historical failure data of PDC drill bits. For example, the average of the composite degradation indexes at which tooth breakage or mechanical drilling speed drops to unacceptable levels occurs historically is used as a threshold, such as 0.8, representing the end of the drill bit's lifespan. Regarding the wear acceleration factor... The settings need to be adjusted based on the actual geological characteristics of the formation. For formations with high abrasiveness and anticipated rapid wear development, It can be set to 2.0 to 2.5 to express a strong nonlinear acceleration effect and ensure that the prediction results are conservative and safe; for conventional or homogeneous strata, It can be set to 1.5 to 2.0 to conform to the general progressive wear pattern. In this embodiment, both versatility and safety are taken into consideration. The preferred setting is 2.0. Regarding preset corrections... This value is only used to prevent calculation overflow during the initial break-in period due to the denominator being close to zero; it is usually taken as a small positive number, such as... This has no substantial impact on the prediction results.

[0078] Figure 4 illustrates the correlation between the composite degradation index and the predicted remaining life of the drill bit. The horizontal axis represents time, the left vertical axis represents the value of the composite degradation index, and the right vertical axis represents the predicted remaining life. The horizontal dashed line corresponds to the preset failure threshold. The figure shows an inverse relationship between the composite degradation index and the predicted remaining life: when the degradation index curve exceeds the failure threshold, the remaining life curve simultaneously drops to a low level, and eventually returns to zero when the degradation index reaches its limit.

[0079] Furthermore, to avoid drill bit jamming or tooth loss accidents due to sudden drill bit failure, this embodiment also introduces a graded early warning mechanism.

[0080] Specifically, a first warning threshold and a second warning threshold are set based on historical failure data. The first warning threshold corresponds to the drill bit entering the initial stage of rapid wear, for example, set to 70% of the failure threshold. The second warning threshold corresponds to the drill bit being on the verge of failure, for example, set to 90% of the failure threshold. This satisfies the numerical relationship that the first warning threshold is less than the second warning threshold, and both are less than the failure threshold. Real-time monitoring of changes in the composite degradation index is conducted: when the composite degradation index is detected to be greater than the first warning threshold, a mild wear warning signal is triggered. At this time, it is recommended to reduce drilling pressure or rotation speed, monitor torque fluctuations, and begin preparing a spare drill bit. When the composite degradation index is detected to be greater than the second warning threshold, a severe wear alarm signal is triggered. This indicates that the remaining life of the drill bit is short, and it is recommended to stop drilling and pull out the drill bit for replacement.

[0081] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining life of a PDC drill bit based on torque time-series data, characterized in that, include: Torque data of the drill bit is collected at a preset sampling frequency to construct torque time series data; Based on torque time series data, wear mode components are extracted using variational mode decomposition algorithm; The variational mode decomposition algorithm is based on a variational objective functional, which includes constraint terms for limiting the bandwidth of the modal components and constraint terms for enhancing the time-domain sparsity of the modal components; the variational objective functional satisfies the following relation: ;in, Based on full set of modal components With the central frequency of the entire set The constructed variational objective functional to be minimized; It is the preset number of modal component decomposition layers; It is the sampling time; It is the first The modal components to be solved at the sampling time The amplitude; It is the first The center frequencies corresponding to the modal components to be solved; It is a balance parameter; It is the L2 norm; It is an L1 norm; Indicates the sampling time Find the partial derivative; It is the Dirac function; It is a temporal convolution operator; The unit is the imaginary unit. Taking any given moment as the current moment, the logarithmic energy feature and multi-scale spectral entropy of the wear mode component are calculated. The logarithmic energy feature and multi-scale spectral entropy are then weighted and fused to obtain the composite degradation index at the current moment. This includes: calculating the coefficient of variation of the logarithmic energy feature within a preset time window to obtain a first coefficient of variation; calculating the coefficient of variation of the multi-scale spectral entropy within a preset time window to obtain a second coefficient of variation; normalizing the first and second coefficients of variation using the sum of the first and second coefficients of variation to obtain the logarithmic energy feature weight and the multi-scale spectral entropy weight; and weighting and summing the logarithmic energy feature and the multi-scale spectral entropy feature using the logarithmic energy feature weight and the multi-scale spectral entropy weight to obtain the composite degradation index. The logarithmic energy feature satisfies the following relationship: ;in, The wear mode component at the sampling time Logarithmic energy characteristics; The wear mode component at the sampling time The amplitude; It is the length of the integration window; It is the initial reference energy; It is an integral variable; It is a logarithmic function; the acquisition of multi-scale spectral entropy includes: performing probability statistics on the wear mode components to obtain their energy probability distribution at several preset frequency scales, calculating the entropy value of the energy probability distribution based on Shannon information entropy, and obtaining multi-scale spectral entropy; calculating the ratio of the composite degradation index to the cumulative running time corresponding to the current moment, multiplying the ratio by a preset wear acceleration factor to obtain the instantaneous degradation rate; obtaining the difference between the composite degradation index and a preset failure threshold, dividing the difference by the instantaneous degradation rate to obtain the remaining life of the PDC drill bit.

2. The PDC drill bit remaining life prediction method based on torque time series data according to claim 1, characterized in that, Solving the variational objective functional includes: converting the variational objective functional into an unconstrained optimization problem using the alternating direction multiplier method, performing iterative calculations in the frequency domain, and filtering several modal components from the iterative calculations to obtain the wear modal components.

3. The PDC drill bit remaining life prediction method based on torque time series data according to claim 2, characterized in that, The step of filtering the several modal components calculated in the iterative calculation to obtain the wear modal components includes: calculating the kurtosis value of the several modal components and selecting the modal component with the largest kurtosis value as the target wear modal component.

4. The PDC drill bit remaining life prediction method based on torque time series data according to claim 1, characterized in that, The remaining lifetime satisfies the following relationship: ;in, It is the time when the drill bit is sampling The remaining lifespan; It is a preset failure threshold; Sampling time The composite degradation index; From the start point of data acquisition to the sampling time The cumulative time; It is a wear-accelerating factor; This is a preset correction option.

5. The PDC drill bit remaining life prediction method based on torque time series data according to claim 1, characterized in that, The step of acquiring torque data of the drill bit at a preset sampling frequency to construct torque time-series data includes: acquiring torque signals using a high-frequency torque sensor; setting a sampling frequency and, according to the Nyquist sampling theorem, using a low-pass filter to remove high-frequency electromagnetic noise in the torque signal that is greater than half the sampling frequency; extracting stable drilling stage data of a preset duration and arranging them according to the order of sampling times to obtain torque time-series data.

6. The PDC drill bit remaining life prediction method based on torque time series data according to claim 1, characterized in that, The method further includes a tiered early warning system, which includes: setting a first early warning threshold and a second early warning threshold, wherein the first early warning threshold is less than the second early warning threshold and both are less than a preset failure threshold; when the composite degradation index is greater than the first early warning threshold, a mild wear warning signal is generated; when the composite degradation index is greater than the second early warning threshold, a severe wear alarm signal is generated and it is recommended to replace the drill bit.

Images