Method for detecting broken line of control loop

Abstract

The invention discloses a method for detecting disconnection of a control loop. The method comprises the following steps: acquiring an original signal data set, performing sampling frequency optimization on the original signal data set to obtain an optimized sampling frequency value, performing resampling and noise processing, and generating a quality evaluation index; performing enhancement processing on the preprocessed signal, extracting time-frequency domain features, performing correlation analysis and time sequence decomposition, and generating a reliable feature set and a reliability report; performing feature space mapping and hierarchical feature extraction based on the reliable feature set, executing fault recognition to obtain fault probability distribution, performing credibility evaluation, and generating an anomaly evaluation report and a final diagnosis result; and extracting time sequence characteristics according to the abnormal assessment report and the diagnosis result, performing trend prediction and risk assessment, and generating an optimization decision scheme. According to the invention, through multi-level signal processing and feature analysis and in combination with a credibility evaluation mechanism, accurate detection and predictive maintenance of the control loop broken line fault are realized.

Description

Technical Field

[0001] The invention belongs to a control system, in particular to a method for detecting disconnection of a control loop. Background Art

[0002] Control loop disconnection detection has important research significance in the field of industrial automation. As the core component of industrial control systems, the integrity and reliability of control loops directly affect the safety and efficiency of the entire production process. Especially in key areas such as continuous production processes, chemical industries, and energy facilities, control loop disconnection failures may lead to system loss of control and production interruptions. Therefore, the development of efficient and reliable disconnection detection methods can not only improve the operational reliability of the system and reduce maintenance costs, but also provide important technical support for predictive maintenance, which is of great significance to improving the level of industrial automation.

[0003] At present, control loop disconnection detection mainly adopts traditional methods such as resistance measurement method, current monitoring method and signal injection method. The resistance measurement method determines the line status by periodically measuring the loop resistance value, but it needs to temporarily interrupt the normal working signal; the current monitoring method detects the disconnection by real-time monitoring the change of the loop current value, but it is easily disturbed by load changes and environmental noise; the signal injection method detects the disconnection by injecting a specific test signal into the loop, but it may affect the transmission of normal control signals. In addition, although the detection method based on neural network has certain fault identification ability, it is difficult to obtain training data and the model generalization ability is limited.

[0004] The existing control loop disconnection detection methods have the following technical problems: First, under complex working conditions, the traditional single signal analysis method is difficult to accurately distinguish disconnection faults from other interference factors, especially when the load changes dynamically, it is easy to have false alarms and missed alarms; second, the existing feature extraction methods mainly focus on static features, and the dynamic features and trend features in the development process of disconnection faults are not captured enough, resulting in limited early fault detection capabilities; third, the detection algorithm is not adaptable enough, and it is difficult to automatically adjust the detection parameters and strategies according to different working conditions, which affects the accuracy and reliability of detection; finally, the existing methods generally lack a credibility assessment mechanism for the detection results, and cannot provide reliable confidence support for maintenance decisions. In addition, in complex systems with multiple loops, it is difficult for existing methods to effectively analyze the propagation mechanism and impact range of faults, which limits the system-level fault warning and risk assessment capabilities. Summary of the invention

[0005] The purpose of the invention is to provide a method for detecting a control loop disconnection to solve the above-mentioned problems existing in the prior art.

[0006] Technical solution, a method for detecting disconnection of a control loop, comprising the following steps:

[0007] S1. Obtain an original signal data set, optimize the sampling frequency of the original signal data set to obtain an optimized sampling frequency value; based on the optimized sampling frequency value, resample the original signal data set to obtain a resampled signal set; perform noise processing on the resampled signal set to obtain a preprocessed signal set; perform data quality assessment on the preprocessed signal set to generate a quality assessment index; wherein the original signal data set includes current signals, voltage signals, temperature signals and vibration signals;

[0008] S2. Based on the preprocessed signal set and the quality assessment index, perform signal enhancement processing to obtain an enhanced signal set; extract time domain and frequency domain features from the enhanced signal set to obtain a feature matrix; perform correlation analysis on the feature matrix to obtain an optimized feature set; decompose the optimized feature set according to time series to obtain a time series feature vector; perform stability assessment on the time series feature vector to generate a reliable feature set and a reliability report;

[0009] S3. Based on the reliable feature set and the reliability report, feature space mapping is performed to obtain a mapping feature set; hierarchical feature extraction is performed on the mapping feature set to obtain a hierarchical feature set; based on the hierarchical feature set, fault identification is performed to obtain a fault probability distribution; credibility assessment is performed on the fault probability distribution to obtain a credibility index; based on the credibility index and the fault probability distribution, anomaly identification is performed to obtain anomaly assessment report and final diagnosis results;

[0010] S4. Based on the abnormal assessment report, the final diagnosis result and the pre-stored historical diagnosis database, the time series features are extracted to obtain a time series pattern set; based on the time series pattern set, trend prediction is performed to obtain a prediction result sequence; based on the prediction result sequence, risk assessment is performed to obtain a risk assessment report; based on the risk assessment report and the prediction result sequence, a decision plan is generated to obtain an optimized decision plan.

[0011] Beneficial effects: The present invention ensures the high quality of basic data, realizes the accurate capture of disconnection characteristics, and improves the accuracy of fault diagnosis; it can adaptively adjust its parameters and strategies, continuously improve the accuracy and reliability of detection, while also reducing maintenance costs, improving the reliability and safety of the system, and providing a strong guarantee for the continuity and stability of industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the present invention.

[0013] Figure 2 This is a flow chart of step S1 of the present invention.

[0014] Figure 3 This is a flow chart of step S2 of the present invention.

[0015] Figure 4This is a flow chart of step S3 of the present invention.

[0016] Figure 5 This is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0017] like Figure 1 As shown, the present application proposes a method for detecting a disconnection in a control loop, comprising the following steps:

[0018] S1. Obtain an original signal data set, optimize the sampling frequency of the original signal data set to obtain an optimized sampling frequency value; based on the optimized sampling frequency value, resample the original signal data set to obtain a resampled signal set; perform noise processing on the resampled signal set to obtain a preprocessed signal set; perform data quality assessment on the preprocessed signal set to generate a quality assessment index; wherein the original signal data set includes current signals, voltage signals, temperature signals and vibration signals;

[0019] S2. Based on the preprocessed signal set and the quality assessment index, perform signal enhancement processing to obtain an enhanced signal set; extract time domain and frequency domain features from the enhanced signal set to obtain a feature matrix; perform correlation analysis on the feature matrix to obtain an optimized feature set; decompose the optimized feature set according to time series to obtain a time series feature vector; perform stability assessment on the time series feature vector to generate a reliable feature set and a reliability report;

[0020] S3. Based on the reliable feature set and the reliability report, feature space mapping is performed to obtain a mapping feature set; hierarchical feature extraction is performed on the mapping feature set to obtain a hierarchical feature set; based on the hierarchical feature set, fault identification is performed to obtain a fault probability distribution; credibility assessment is performed on the fault probability distribution to obtain a credibility index; based on the credibility index and the fault probability distribution, anomaly identification is performed to obtain anomaly assessment report and final diagnosis results;

[0021] S4. Based on the abnormal assessment report, the final diagnosis result and the pre-stored historical diagnosis database, the time series features are extracted to obtain a time series pattern set; based on the time series pattern set, trend prediction is performed to obtain a prediction result sequence; based on the prediction result sequence, risk assessment is performed to obtain a risk assessment report; based on the risk assessment report and the prediction result sequence, a decision plan is generated to obtain an optimized decision plan.

[0022] like Figure 2 As shown, according to one aspect of the present application, step S1 is further:

[0023] S11, obtaining an original signal data set, calculating the statistical features of each signal within a preset time window, and obtaining a feature vector; normalizing the values ​​in the feature vector to obtain a standardized feature vector; determining a weight coefficient group based on the degree of numerical deviation in the standardized feature vector; multiplying the weight coefficient group by the standardized feature vector to obtain a weighted feature vector;

[0024] S12, based on the preset sampling frequency range, calculate the change rate of the weighted feature vector to obtain the change rate value; calculate the sampling frequency adjustment coefficient according to the change rate value and the current sampling frequency; multiply the sampling frequency adjustment coefficient by the current sampling frequency to obtain a new sampling frequency; determine whether the new sampling frequency is within the sampling frequency range, if within the range, directly use the new sampling frequency as the optimized sampling frequency value; if out of the range, take the boundary value and output the optimized sampling frequency value;

[0025] S13, based on the optimized sampling frequency value, interpolating the original signal data set to obtain an interpolated data set; performing anti-aliasing filtering on the interpolated data set to obtain a filtered data set; performing timed sampling on the filtered data set according to the optimized sampling frequency value to obtain a resampled signal set;

[0026] S14, obtaining a resampled signal set, calculating a local fluctuation feature of the signal, and obtaining a fluctuation feature set; determining a filtering parameter based on the fluctuation feature set, and obtaining a filtering parameter set; and using the filtering parameter set to adaptively filter the resampled signal set, and obtaining a preprocessed signal set;

[0027] S15. Based on the preprocessed signal set, the signal-to-noise ratio of the signal is calculated to obtain a signal-to-noise ratio value group; based on the preprocessed signal set, the continuity index of the signal is calculated to obtain a continuity index group; based on the preprocessed signal set, the integrity index of the signal is calculated to obtain an integrity index group; the signal-to-noise ratio value group, the continuity index group and the integrity index group are normalized and weighted to obtain a quality assessment index.

[0028] In one embodiment of the present application, the original current signal i(t), voltage signal v(t), temperature signal T(t) and vibration signal V(t) are obtained, and the characteristic vector F of each signal in the time window w is calculated; for the current signal: extract the instantaneous value ia, mean im, standard deviation is, and change rate ir; for the voltage signal: extract the instantaneous value va, mean vm, standard deviation vs, and change rate vr; for the temperature signal: extract the mean Tm and change rate Tr; for the vibration signal: extract the mean Vm and change rate Vr; output the characteristic vector F=[ia, im, is, ir, va, vm, vs, vr, Tm, Tr, Vm, Vr].

[0029] Based on the eigenvector F and the preset sampling frequency range [fmin, fmax], a new adaptive function A(F) is introduced: A(F)=α·(w1·ΔF / Δt + w2·σF + w3·|F - Fref|); where ΔF / Δt is the rate of change of the eigenvector, σF is the volatility of the eigenvector, |F - Fref| is the deviation from the reference value, w1, w2, w3 are adaptive weight coefficients, and α is the correction factor. Calculate the new sampling frequency fnew = fbase·(1 + A(F)); where fbase is the base sampling frequency, ensure that the new sampling frequency fnew is within the range [fmin, fmax], and output the optimized sampling frequency fnew.

[0030] Based on the feature vector F and the historical database H, the information entropy En = -Σ(pi·log(pi)) is calculated through the dynamic entropy weight threshold method (DEWT); where pi is the probability distribution of the i-th feature, and the dynamic threshold matrix T = [t1, t2, ..., tn] ti = μi + k·σi·(1-Eni); where μi is the mean of the i-th feature, σi is the standard deviation, Eni is the corresponding information entropy, k is the adjustable coefficient, and the dynamic threshold matrix T is output.

[0031] Based on the original signal set S, dynamic threshold matrix T and sampling frequency fnew, the adaptive wavelet packet decomposition and reconstruction (AWPR) method is used to perform multi-scale wavelet packet decomposition; coefficients are screened according to the dynamic threshold matrix T; based on the sampling frequency fnew, the signal is reconstructed and the processed signal set S' is output. Based on the processed signal set S', data quality assessment is performed and quality assessment indicators are output.

[0032] This embodiment achieves high-quality preprocessing of the original signal through multi-level signal processing and optimization strategies. First, through the sampling frequency optimization mechanism, the sampling frequency can be adaptively adjusted, which ensures that the key features of the signal can be accurately captured under different working conditions without losing important information. In the resampling process, interpolation algorithms and anti-aliasing filtering techniques are used to effectively reduce signal distortion during the sampling process. By performing adaptive noise processing on the resampled signal, the signal-to-noise ratio of the signal is improved while maintaining the original characteristics of the signal. Finally, through a quality evaluation mechanism that comprehensively considers multi-dimensional indicators such as signal-to-noise ratio, continuity and integrity, a reliable quality assurance is provided for subsequent signal processing. This embodiment not only improves the quality of the signal, but also lays a solid foundation for subsequent feature extraction and fault diagnosis, and improves the accuracy and reliability of disconnection detection.

[0033] According to one aspect of the present application, step S11 is further:

[0034] S111, obtaining an original signal data set, performing block processing on each signal within a preset time window to obtain a signal block sequence; calculating the mean of each signal block sequence to obtain a mean sequence; calculating the standard deviation of each signal block sequence to obtain a standard deviation sequence; calculating the peak value of each signal block sequence to obtain a peak value sequence; combining the mean sequence, the standard deviation sequence and the peak value sequence to obtain a single signal feature vector;

[0035] S112, obtaining the current signal and the voltage signal in the original signal data set, calculating the phase difference sequence, and obtaining the phase feature vector; obtaining the temperature signal in the original signal data set, and calculating the thermal response feature based on the temperature signal and the current signal to obtain the thermal feature vector; obtaining the vibration signal in the original signal data set, and calculating the coupling feature based on the vibration signal and the voltage signal to obtain the coupling feature vector; combining the phase feature vector, the thermal feature vector, and the coupling feature vector to obtain the combined feature vector;

[0036] S113, based on the single signal feature vector, calculating the rate of change of the feature to obtain a rate of change sequence; based on the combined feature vector, calculating the trend of the feature to obtain a trend sequence; based on the rate of change sequence and the trend sequence, constructing a time series matrix to obtain a time series feature vector;

[0037] S114. Based on the single signal feature vector, the combined feature vector and the time series feature vector, the covariance matrix of each feature vector is calculated to obtain a correlation matrix; based on the correlation matrix, feature combination is performed to obtain a feature vector; the feature vector is normalized to obtain a standardized feature vector; the discrete degree of each component in the standardized feature vector is calculated to obtain a weight coefficient group; the weight coefficient group is multiplied by the standardized feature vector to obtain a weighted feature vector.

[0038] This embodiment achieves comprehensive capture and optimization of the original signal features through multi-dimensional feature extraction and fusion strategies. First, by performing block processing within a fixed time window, the local statistical characteristics of the signal, including the mean, standard deviation and peak value, etc., can be captured. These characteristics can effectively reflect the basic characteristics and change trends of the signal. Secondly, by analyzing the relationship between multiple signals, including the phase difference between current and voltage, the thermal response characteristics of current and temperature, and the coupling characteristics of voltage and vibration, the interaction and dependency between various physical quantities in the control loop can be fully captured. Thirdly, by calculating the rate of change and trend of the features, a time series feature vector is constructed, so that the system can effectively capture the dynamic characteristics of the signal. Finally, through the correlation analysis and weight optimization of the features, the adaptive fusion of the features is achieved, which not only ensures the integrity of the features, but also reduces the redundancy of the data. This embodiment improves the accuracy and reliability of subsequent disconnection detection.

[0039] According to one aspect of the present application, step S13 is further:

[0040] S131, based on the original signal data set and the optimized sampling frequency value, calculate the sampling point spacing to obtain a spacing sequence; based on the spacing sequence, perform segmentation processing to obtain a segment sequence; based on the segment sequence, calculate the local features of each segment of data to obtain a feature sequence; select the optimal interpolation algorithm according to the feature sequence to obtain an interpolation data set;

[0041] S132, based on the interpolation data set, calculating the signal spectrum distribution to obtain a spectrum sequence; based on the spectrum sequence, constructing filter parameters to obtain a filter parameter group; applying the filter parameter group to perform bandpass filtering to obtain a bandpass signal sequence; performing phase correction on the bandpass signal sequence to obtain a filtered data set;

[0042] S133. Obtain a filtered data set, calculate the sampling moment according to the optimized sampling frequency value, and obtain a sampling moment sequence; calculate the signal value of each sampling moment sequence to obtain a sampling value sequence; based on the sampling value sequence, perform a sampling consistency check to obtain a consistency index sequence; perform sampling correction according to the consistency index sequence to obtain a resampled signal set.

[0043] This embodiment achieves high-quality signal reconstruction and optimization through a signal resampling strategy. First, the system ensures the complete retention of signal characteristics during the resampling process by calculating the optimal sampling point spacing and segmented processing. Secondly, through spectrum analysis and adaptive filter design, the aliasing effect and noise interference can be effectively suppressed to improve the quality of the signal. Thirdly, through phase correction and sampling consistency check, the timing accuracy and continuity of the resampled signal are guaranteed. This embodiment not only ensures the integrity and accuracy of the signal, but also improves the system's adaptability to various complex working conditions. Through precise control and consistency check of the sampling moment, the characteristic information of the disconnection fault can be accurately captured, providing reliable data support for subsequent fault diagnosis.

[0044] According to one aspect of the present application, step S14 is further:

[0045] S141. Obtain the resampled signal set R, segment the signal and obtain the segment sequence Sq; calculate the local variance of each signal segment to obtain the variance sequence Sv; calculate the jump point position of each signal segment to obtain the jump position sequence Jp; calculate the fluctuation period of each signal segment to obtain the period sequence Cp; combine the variance sequence Sv, the jump position sequence Jp and the period sequence Cp to obtain the fluctuation feature set V.

[0046] S142. Obtain the fluctuation feature set V, calculate the basic filter bandwidth based on the variance sequence Sv, and obtain the bandwidth parameter sequence Bw; determine the filter window length according to the jump position sequence Jp, and obtain the window length sequence Wl; calculate the filter order based on the period sequence Cp, and obtain the filter order sequence Or; combine the bandwidth parameter sequence Bw, the window length sequence Wl and the filter order sequence Or to obtain the initial parameter set Pi.

[0047] S143. Obtain the resampled signal set R and the initial parameter set Pi, perform segmented filtering on the signal, and obtain an initial filtering sequence F1; calculate the filtering effect of the initial filtering sequence F1 to obtain an effect evaluation value E1; fine-tune the initial parameter set Pi according to the effect evaluation value E1 to obtain an adjustment parameter set Pa; use the adjustment parameter set Pa to perform secondary filtering to obtain a secondary filtering sequence F2.

[0048] S144, obtain the secondary filtering sequence F2, calculate the residual sequence of the signal, and obtain the residual sequence Re; perform spectrum analysis on the residual sequence Re to obtain the spectrum feature vector Sf; optimize the filtering parameters according to the spectrum feature vector Sf to obtain the filtering parameter set Fp; use the filtering parameter set Fp to perform final filtering on the resampled signal set R to obtain the preprocessed signal set P.

[0049] This embodiment achieves efficient noise reduction and optimization processing of resampled signals through an adaptive filtering strategy. First, by segmenting the signal and calculating features such as local variance, jump point position, and fluctuation period, the local fluctuation characteristics of the signal can be accurately captured, and these features provide an important basis for subsequent adaptive filtering. Secondly, the key parameters of the filter, including bandwidth, window length, and filter order, are automatically determined based on the fluctuation characteristics, so that the filter can adaptively adjust its performance according to the characteristics of the signal. Thirdly, through two-stage iterative filtering processing and effect evaluation, the filtering parameters can be continuously optimized to ensure the best filtering effect. Finally, the accuracy of filtering is further improved through residual analysis and spectrum feature optimization. This embodiment can not only effectively suppress various types of noise interference, but also maintain the important characteristics of the signal, and improve the sensitivity and reliability of disconnection detection.

[0050] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0051] S21. Based on the quality assessment index, calculate the enhancement coefficient to obtain an enhancement coefficient group; perform a time domain enhancement operation on the enhancement coefficient group and the preprocessed signal set to obtain time domain enhancement data; perform a frequency domain enhancement operation on the enhancement coefficient group and the preprocessed signal set to obtain frequency domain enhancement data; fuse the time domain enhancement data and the frequency domain enhancement data to obtain an enhanced signal set;

[0052] S22, based on the enhanced signal set, calculating the first order difference of the signal to obtain a first difference sequence; based on the first difference sequence, calculating the second order difference of the signal to obtain a second difference sequence; extracting time domain features according to the first difference sequence and the second difference sequence to obtain a time domain feature sub-matrix; performing spectrum analysis on the enhanced signal set to obtain a frequency domain feature sub-matrix; merging the time domain feature sub-matrix and the frequency domain feature sub-matrix to obtain a feature matrix;

[0053] S23, based on the feature matrix, calculating the correlation coefficient between each column of the matrix to obtain a correlation coefficient matrix; according to the correlation coefficient matrix, calculating the importance score of each feature to obtain an importance score vector; according to the importance score vector, performing feature selection on the feature matrix to obtain an optimized feature set;

[0054] S24, based on the optimized feature set, segmenting the feature data according to a preset fixed time window to obtain a time series group; extracting a trend component from the time series group to obtain a trend feature vector; extracting a period component from the time series group to obtain a period feature vector; extracting a mutation component from the time series group to obtain a mutation feature vector; combining the trend feature vector, the period feature vector and the mutation feature vector to obtain a time series feature vector;

[0055] S25. Based on the time series feature vector and the preprocessed signal set, calculate the time stability index of the feature to obtain a stability index group; based on the time series feature vector and the preprocessed signal set, calculate the anti-interference ability index of the feature to obtain an anti-interference index group; filter the features according to the stability index group and the anti-interference index group to obtain a reliable feature set; generate a reliability report based on the evaluation data of the screening process.

[0056] In one embodiment of the present application, based on the optimized signal set S' and the data quality indicator Q, the adaptive feature enhancement projection (AFEP) method is used to construct an enhancement matrix E = [Et, Ef]; wherein Et is the time domain feature matrix, and Ef is the frequency domain feature matrix; the enhancement coefficient is calculated: λ = f(Q)·(γt·Et + γf·Ef); wherein f(Q) is an adjustment function based on data quality; γt, γf are time-frequency domain weight coefficients; feature enhancement is performed: S'' = S'·λ, and the enhanced signal set S'' is output.

[0057] Based on the enhanced signal set S'', the first-order eigenvector F1 is calculated by the hierarchical eigendecomposition (HFD) method: current mutation rate ri = Δi / Δt; voltage response rate rv = Δv / Δt; impedance change rate rz = Δz / Δt; the second-order eigenvector F2 is calculated: current-voltage phase difference θiv; power factor change rate rpf; energy loss rate rel; a combined feature matrix C = [F1, F2] is constructed, and the feature matrix C is output. Among them, Δi is the change of current, Δt is the change of time, Δv is the change of voltage, and Δz is the change of impedance.

[0058] Based on the feature matrix C, the dynamic mutual information network (DMIN) algorithm is used to calculate the mutual information matrix I between features: I(x, y) = ∑P(x, y)·log[P(x, y) / (P(x)·P(y))]; where P(x, y) is the joint probability distribution; P(x) and P(y) are the marginal probability distributions; a feature association network G is constructed, the maximum weight path Pmax is extracted, and the optimized feature set Copt is output.

[0059] Based on the optimized feature set Copt, the time window sequence W = [w1, w2, ..., wn] is set through the recursive pattern decomposition (RPD) method; for each window, the following are calculated: trend feature Ft; period feature Fp; mutation feature Fm; synthetic time series feature vector Ts = [Ft, Fp, Fm], and the time series feature vector Ts is output.

[0060] Based on the time series feature vector Ts and the optimized feature set Copt, the feature stability evaluation (FSE) method is used to calculate the feature stability index S: S = α1·Rc + α2·Rt + α3·Rd; where Rc is the feature consistency, Rt is the time stability, and Rd is the anti-interference ability; α1, α2, α3 are weight coefficients; a reliability report R is generated; and finally the reliable feature set F_reliable and the reliability report R are output.

[0061] This embodiment adopts a signal enhancement and feature extraction strategy to achieve accurate capture of the control loop disconnection characteristics. Through dual enhancement processing in the time domain and frequency domain, the key features in the signal can be effectively highlighted while suppressing interference components. In the feature extraction process, a comprehensive time-frequency feature description is obtained by calculating the first-order and second-order differences and combining spectrum analysis. Through correlation analysis and feature selection, the most diagnostically valuable features can be automatically identified and retained, reducing data redundancy. Through time series decomposition and stability evaluation, not only the temporal continuity of the features is guaranteed, but also the anti-interference ability of the features is ensured. This embodiment improves the sensitivity and stability of disconnection detection while reducing the false alarm rate.

[0062] According to one aspect of the present application, step S22 is further:

[0063] S221. Based on the enhanced signal set, calculate the local extreme value points of the signal to obtain an extreme value sequence; based on the extreme value sequence, calculate the signal envelope to obtain a first envelope sequence; calculate the first-order difference of the enhanced signal set to obtain a first difference sequence; calculate the second-order difference of the first difference sequence to obtain a second difference sequence; calculate the statistical features of the first difference sequence and the second difference sequence to obtain a time domain statistical sequence; combine the first envelope sequence and the time domain statistical sequence to obtain a basic time domain feature vector;

[0064] S222, based on the enhanced signal set, performing short-time Fourier transform to obtain a time-frequency spectrum matrix; performing energy distribution analysis on the time-frequency spectrum matrix to obtain an energy distribution sequence; calculating the frequency center drift of the time-frequency spectrum matrix to obtain a center drift sequence; combining the energy distribution sequence and the center drift sequence to obtain a frequency domain basic feature vector;

[0065] S223, based on the basic time domain feature vector and the frequency domain basic feature vector, calculate the phase relationship between the features to obtain a phase relationship matrix; based on the basic time domain feature vector and the frequency domain basic feature vector, calculate the time-frequency coupling degree of the features to obtain a coupling degree sequence; based on the phase relationship matrix and the coupling degree sequence, construct a joint feature to obtain a time-frequency joint feature vector;

[0066] S224. Based on the basic time domain feature vector, grouping is performed according to feature correlation to obtain a time domain feature sub-matrix; based on the frequency domain basic feature vector and the time-frequency joint feature vector, the main feature components are extracted to obtain a frequency domain feature sub-matrix; feature standardization is performed on the time domain feature sub-matrix and the frequency domain feature sub-matrix to obtain a standardized feature group; the standardized feature group is reorganized according to a preset format to obtain a feature matrix.

[0067] This embodiment achieves comprehensive capture and feature optimization of disconnection features through a comprehensive time-frequency domain feature extraction strategy. First, by calculating the local extreme points and envelope features of the signal, combined with first-order and second-order differential analysis, the time domain change characteristics of the signal are fully captured. Secondly, through short-time Fourier transform and energy distribution analysis, the frequency domain characteristics of the signal can be accurately obtained, including key information such as energy distribution and frequency center drift. Thirdly, by calculating the phase relationship and time-frequency coupling between features, a complete time-frequency joint feature is established, which effectively reflects the dynamic characteristics of the disconnection fault. Finally, through the standardized processing and reorganization optimization of the features, a highly integrated feature matrix is ​​formed. This embodiment not only improves the feature discrimination ability, but also enhances the system's ability to recognize various complex fault modes.

[0068] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0069] S31. Based on the reliability report, calculate the feature weights to obtain a weight matrix; perform mapping transformation on the weight matrix and the reliable feature set to obtain an initial mapping set; perform dimensional optimization on the initial mapping set to obtain a mapping feature set;

[0070] S32, based on the mapping feature set, extracting mutation feature parameters to obtain a mutation feature subset; based on the mapping feature set, extracting fluctuation feature parameters to obtain a fluctuation feature subset; based on the mapping feature set, extracting trend feature parameters to obtain a trend feature subset; combining the mutation feature subset, the fluctuation feature subset and the trend feature subset to obtain a hierarchical feature set;

[0071] S33, based on the hierarchical feature set, calculating the time domain identification index to obtain a time domain index group; based on the hierarchical feature set, calculating the frequency domain identification index to obtain a frequency domain index group; based on the hierarchical feature set, calculating the power domain identification index to obtain a power index group; performing probability conversion on the time domain index group, the frequency domain index group and the power index group to obtain a fault probability distribution;

[0072] S34, based on the fault probability distribution and the pre-stored historical accuracy data, calculate the evidence strength index to obtain the evidence strength value; obtain the current working condition data, calculate the working condition similarity, and obtain the similarity value; calculate the comprehensive credibility according to the evidence strength value and the similarity value to obtain the credibility index;

[0073] S35. Based on the credibility index, fault probability distribution and pre-configured normal operating condition model, calculate the deviation between the current state and the normal model to obtain the deviation index; calculate the degree of abnormality according to the deviation index and the credibility index to obtain the abnormality assessment report; determine the final diagnosis result based on the abnormality assessment report and the fault probability distribution.

[0074] In one embodiment of the present application, based on a reliable feature set F_reliable and a reliability report R, a dynamic feature fusion mapping (DFFM) method is adopted to construct a mapping matrix M: M = Wd·Fd + Wt·Ft + Wn·Fn; wherein Fd is a deterministic feature subset, Ft is a time-varying feature subset, and Fn is a noise-related feature subset; Wd, Wt, Wn are corresponding weight matrices; a feature space transformation is performed: F_new = F_reliable·M; the compression rate Δ= g(R) is calculated by an adaptive feature compression (AFC) algorithm, wherein g(R) is a mapping function based on the reliability report, and the optimized feature space F_opt is output.

[0075] Based on the optimized feature space F_opt, the progressive feature layering (PFL) method is used to extract the first-level features (complete disconnection features): current drop index Ic; voltage mutation index Vc; power loss index Pc; the second-level features (semi-disconnection features): contact resistance fluctuation Rw; temperature gradient change Tg; vibration feature change Vf; the third-level features (virtual connection features): intermittent fault index If; nonlinear impedance feature Zn; thermal cycle feature Tc; and output a multi-level feature set F_multi.

[0076] Based on the multi-level feature set F_multi, the hybrid domain fault recognition (HDFR) algorithm is adopted to construct the identification matrix D: D= [Dt | Df | Dp]; where Dt is the time domain identification sub-matrix, Df is the frequency domain identification sub-matrix, and Dp is the power domain identification sub-matrix; the fault probability vector P is calculated: P = σ(D·F_multi); where σ is the improved Softmax function: σ(x)i=exi·(1+βi) / Σ(exj·(1+βj)); βi is the reliability weighting factor, exi represents the exponential operation result of the natural number e and xi; the fault type probability distribution P is output.

[0077] Based on the probability distribution P of fault types, the evidence strength E is calculated by the multidimensional credibility evaluation (MDCE) method: E = h(P, H, C); where H is the historical diagnosis accuracy, C is the current working condition similarity, and h is the evidence fusion function; the credibility index T is generated: T = κ·E + (1-κ)·S; where S is the signal quality index and κ is the balance coefficient; the credibility T of the diagnosis result is output.

[0078] Based on the credibility T of the diagnosis result and the probability distribution P of the fault type, the dynamic anomaly detection (DAD) method is adopted to construct the normal operating condition model N, and calculate the deviation index: d = ||P - N||·(1 + λ·T); where λ is the credibility influencing factor, generate the anomaly level assessment A, output the anomaly assessment report A and the final diagnosis result D_final.

[0079] This embodiment achieves accurate line break fault diagnosis through highly integrated feature mapping and multi-level fault identification strategies. First, through feature space mapping, the original features are converted into a more discriminative feature space, which improves the separability of the features. Through hierarchical feature extraction, fault features can be captured from different levels, including mutation features, fluctuation features, and trend features. In the fault identification process, through multi-dimensional analysis in the time domain, frequency domain, and power domain, combined with a credibility assessment mechanism, line break faults can be accurately identified while effectively reducing the misjudgment rate. This embodiment not only improves the accuracy of diagnosis, but also enhances the robustness of the system, and can adapt to various complex working conditions.

[0080] According to one aspect of the present application, step S32 is further:

[0081] S321, based on the mapping feature set, calculating the instantaneous change rate of the signal to obtain a change rate sequence; performing threshold segmentation on the change rate sequence to obtain a segmented sequence; based on the segmented sequence, calculating the jump amplitude of each segment of the signal to obtain an amplitude sequence; based on the segmented sequence, calculating the jump time interval between adjacent segments to obtain a time series; combining the amplitude sequence and the time series to obtain a mutation feature subset;

[0082] S322, based on the mapping feature set, calculate the local fluctuation period to obtain a period sequence; based on the period sequence, calculate the fluctuation envelope to obtain a second envelope sequence; perform frequency decomposition on the second envelope sequence to obtain a frequency sequence; based on the second envelope sequence, calculate the energy distribution of the fluctuation to obtain an energy sequence; combine the frequency sequence and the energy sequence to obtain a fluctuation feature subset;

[0083] S323, based on the mapping feature set, applying adaptive polynomial fitting to obtain a fitting coefficient sequence; based on the fitting coefficient sequence, calculating a fitting residual sequence; based on the fitting residual sequence, extracting the trend slope to obtain a slope sequence; based on the slope sequence, calculating the acceleration feature of the trend to obtain an acceleration sequence; combining the slope sequence and the acceleration sequence to obtain a trend feature subset;

[0084] S324. Based on the mutation feature subset, the fluctuation feature subset and the trend feature subset, the correlation between the feature subsets is calculated to obtain a correlation matrix; based on the correlation matrix, the feature importance is sorted to obtain an importance sequence; according to the importance sequence, the features are screened and reorganized to obtain a reorganized feature matrix; the reorganized feature matrix is ​​organized into a hierarchical structure to obtain a hierarchical feature set.

[0085] This embodiment realizes the accurate decomposition and hierarchical expression of the disconnection fault features through a hierarchical feature extraction strategy. First, by calculating the instantaneous rate of change and jump characteristics, the mutation characteristics of the disconnection fault are accurately captured. Secondly, by analyzing the local fluctuation period and energy distribution, the fluctuation characteristics of the signal can be accurately described. Thirdly, through adaptive polynomial fitting and trend analysis, the trend characteristics of the fault evolution are obtained. Finally, through feature importance evaluation and hierarchical reorganization, a complete hierarchical feature system is established. This embodiment not only improves the interpretability of the features, but also enhances the system's ability to identify different types of faults, providing reliable feature support for fault diagnosis.

[0086] According to one aspect of the present application, step S33 is further:

[0087] S331. Obtain the hierarchical feature set L, extract the time domain feature sequence, and obtain the time series feature group Ts; calculate the statistical features of the time series feature group Ts to obtain the statistical feature sequence St; analyze the mutation characteristics of the features to obtain the mutation feature sequence Mt; combine the statistical feature sequence St and the mutation feature sequence Mt to obtain the time domain indicator group Dt.

[0088] S332. Obtain the hierarchical feature set L, perform multi-scale spectrum decomposition, and obtain the spectrum decomposition sequence Fd; calculate the energy distribution of each frequency band to obtain the energy distribution sequence Ed; analyze the energy migration between frequency bands to obtain the migration feature sequence Tf; combine the energy distribution sequence Ed and the migration feature sequence Tf to obtain the frequency domain indicator group Df.

[0089] S333. Obtain the hierarchical feature set L, calculate the instantaneous power feature, and obtain the power feature sequence Ps; analyze the power fluctuation law to obtain the fluctuation feature sequence Os; calculate the power loss feature to obtain the loss feature sequence Ls; combine the fluctuation feature sequence Os and the loss feature sequence Ls to obtain the power index group Dp.

[0090] S334. Obtain the time domain indicator group Dt, the frequency domain indicator group Df and the power indicator group Dp, calculate the confidence of each indicator, and obtain the confidence sequence Cs; perform feature weighting based on the confidence sequence Cs to obtain the weight sequence Ws; perform multi-feature probability mapping to obtain the probability mapping sequence Ms; normalize the probability mapping sequence Ms to obtain the fault probability distribution P.

[0091] This embodiment realizes comprehensive probability evaluation and feature mapping of line break faults through a multi-domain joint analysis strategy. First, in the time domain analysis, by extracting timing features and statistical features and combining mutation characteristic analysis, the time domain behavior characteristics of the signal are fully captured. Secondly, in the frequency domain analysis, through multi-scale spectrum decomposition and energy distribution analysis, the energy migration characteristics between frequency bands can be accurately identified, and these characteristics are of great significance for the early detection of line break faults. Thirdly, in the power domain analysis, by calculating the instantaneous power characteristics and power loss characteristics, the abnormal power fluctuations caused by line break faults can be effectively identified. Finally, through feature confidence evaluation and multi-feature probability mapping, a reliable fault probability distribution model is established. This embodiment not only improves the accuracy of fault identification, but also enhances the anti-interference ability and adaptability of the system.

[0092] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0093] S41. Based on the abnormal assessment report, the final diagnosis result and the pre-stored historical diagnosis database, extract the historical diagnosis sequence features and the current diagnosis features to obtain the historical feature sequence and the current feature sequence; calculate the temporal correlation between the historical feature sequence and the current feature sequence to obtain a correlation matrix; based on the correlation matrix, extract the key evolution features to obtain a temporal pattern set;

[0094] S42, based on the time series pattern set, construct a state prediction model to obtain a prediction model parameter group; according to the prediction model parameter group, calculate the future state sequence to obtain a prediction state sequence; calculate the confidence interval of the prediction state sequence to obtain an interval boundary sequence; combine the prediction state sequence and the interval boundary sequence to obtain a prediction result sequence;

[0095] S43. Based on the prediction result sequence, the risk value of each subsystem is calculated to obtain a subsystem risk group; based on the subsystem risk group, the overall risk of the system is calculated to obtain a system risk value; based on the system risk value, the current risk level is determined to obtain a risk level matrix; the system risk value and the risk level matrix are integrated to obtain a risk assessment report;

[0096] S44. Based on the risk assessment report and the prediction result sequence, a set of optional decision plans is generated to obtain a decision plan set; the safety score of each plan in the decision plan set is calculated to obtain a safety score group; the cost score of each plan in the decision plan set is calculated to obtain a cost score group; the timeliness score of each plan in the decision plan set is calculated to obtain a timeliness score group; based on a comprehensive evaluation of the safety score group, the cost score group and the timeliness score group, an optimized decision plan is obtained.

[0097] In one embodiment of the present application, based on the historical diagnosis database H and the current diagnosis result D_final, the incremental temporal pattern mining (ISPM) method is adopted to construct a temporal feature matrix Ts: Ts(t) = [Dh(t), Dc(t), Df(t)]; wherein Dh(t) is the historical diagnosis sequence, Dc(t) is the current diagnosis result, and Df(t) is the predicted feature sequence; the temporal correlation matrix R is calculated: R = ψ(Ts, Δt), wherein ψ is the temporal correlation function and Δt is the time window; the key evolution pattern P is extracted, and the temporal pattern set P_set is output.

[0098] Based on the time series pattern set P_set, the prediction model is constructed through the dynamic trend prediction (DTP) algorithm: Y(t+τ) =φ(X(t), θ(t)); where X(t) is the current state vector, θ(t) is the dynamic parameter matrix, τ is the prediction time domain, and φ is the adaptive prediction function; the reliability interval is calculated: [Ymin, Ymax] = Y(t+τ)±η·σ(t); where σ(t) is the prediction standard deviation, η is the credibility coefficient, and the prediction result sequence Y_pred is output.

[0099] Based on the prediction result sequence Y_pred, the multi-level risk assessment (MLRA) method is used to calculate the system risk index Sr: Sr = Σ(wi·ri)·f(t); where ri is the risk value of each subsystem, wi is the weight coefficient, and f(t) is the time decay function; the risk level matrix L is generated: L = [Lc | Lt | Le]; where Lc is the current risk level, Lt is the trend risk level, and Le is the environmental risk level; the risk assessment report R_risk is output.

[0100] Based on the risk assessment report R_risk and the prediction result Y_pred, the decision space D is constructed through the adaptive decision optimization (ADO) algorithm: D = {d1, d2, ..., dn}, where di is an optional decision plan; the optimal decision is calculated: d_opt = argmax(U(di)); where U(di) = α·S(di) + β·C(di) + γ·T(di), S(di) is the safety score, C(di) is the cost score, and T(di) is the timeliness score; α, β, γ are weight coefficients, and the optimized decision plan d_opt is output.

[0101] This embodiment achieves the prediction and optimization decision of disconnection faults through time series analysis and risk assessment mechanism. First, by extracting time series features and pattern recognition, an accurate trend prediction model is established. Based on the prediction results, potential risks can be evaluated in a timely manner, and corresponding risk levels and early warning information can be generated. By comprehensively considering multiple dimensions such as safety, cost and timeliness, the optimal decision plan can be automatically generated. This embodiment can not only prevent the occurrence of disconnection faults, but also optimize maintenance plans, reduce maintenance costs and downtime. At the same time, through continuous risk assessment and decision optimization, the system can continuously improve its prediction and decision-making capabilities and realize intelligent preventive maintenance.

[0102] According to one aspect of the present application, step S42 is further:

[0103] S421, based on the time series pattern set, calculating the autocorrelation coefficient of the time series data to obtain a correlation coefficient sequence; based on the correlation coefficient sequence, determining the order of the prediction model to obtain the order parameter; according to the order parameter, extracting the key state variables to obtain a state variable sequence; normalizing the state variable sequence to obtain a prediction model parameter group;

[0104] S422, based on the prediction model parameter group and the most recent state variable sequence, calculate the short-term prediction sequence; calculate the error range of the short-term prediction sequence to obtain a short-term error sequence; perform drift correction on the short-term prediction sequence to obtain a corrected sequence; combine the corrected sequence and the short-term error sequence to obtain a short-term prediction result;

[0105] S423, based on the prediction model parameter group and the pre-stored historical trend data, calculate the mid-term trend characteristics to obtain a trend characteristic sequence; based on the trend characteristic sequence, perform progressive prediction to obtain a mid-term prediction value sequence; calculate the uncertainty of the mid-term prediction value sequence to obtain an uncertainty sequence; combine the mid-term prediction value sequence and the uncertainty sequence to obtain a mid-term prediction result;

[0106] S424. Based on the short-term prediction results and the medium-term prediction results, the weight coefficients of each prediction result are calculated to obtain a weight sequence; the short-term prediction results and the medium-term prediction results are fused according to the weight sequence to obtain a prediction state sequence; based on the prediction state sequence, the confidence level of the fused prediction is calculated to obtain an interval boundary sequence; the prediction state sequence and the interval boundary sequence are combined in chronological order to obtain a prediction result sequence.

[0107] This embodiment achieves accurate prediction and uncertainty quantification of disconnection faults through prediction model construction and result fusion strategy. First, by analyzing the autocorrelation characteristics of time series data, the optimal order of the prediction model is adaptively determined, thereby improving the prediction accuracy of the model. Secondly, through short-term prediction and error analysis, the recent fault risk can be accurately predicted, and the reliability of the prediction can be improved through drift correction. Thirdly, through medium-term trend feature analysis and progressive prediction, the fault trend prediction of a longer time scale is achieved. Finally, through weighted fusion of prediction results and confidence interval calculation, a complete sequence of prediction results is provided. This embodiment not only improves the accuracy of the prediction, but also provides a reliable uncertainty estimate for the prediction results.

[0108] According to one aspect of the present application, step S43 is further:

[0109] S431. Obtain the prediction result sequence Y, extract the characteristics of each subsystem, and obtain the subsystem characteristic sequence Sf; calculate the characteristic deviation to obtain the deviation sequence Dv; analyze the characteristic change trend to obtain the trend sequence Ts; calculate the risk index based on the deviation sequence Dv and the trend sequence Ts to obtain the subsystem risk group Sr.

[0110] S432. Obtain the subsystem risk group Sr, calculate the risk propagation coefficient, and obtain the propagation coefficient sequence Pc; analyze the coupling relationship between the subsystems to obtain the coupling matrix Cm; calculate the overall risk based on the propagation coefficient sequence Pc and the coupling matrix Cm to obtain the system risk value Rs.

[0111] S433. Obtain the system risk value Rs, calculate the risk threshold sequence, and obtain the threshold sequence Tv; compare the system risk value Rs with the threshold sequence Tv to obtain the risk rating sequence Gr; analyze the risk evolution law to obtain the evolution feature sequence Ef; combine the risk rating sequence Gr and the evolution feature sequence Ef to obtain the risk level matrix L.

[0112] S434. Obtain the system risk value Rs and the risk level matrix L, generate a risk trend graph, and obtain the trend data sequence Td; calculate key risk indicators to obtain the indicator sequence Ki; analyze risk control suggestions to obtain the suggestion sequence As; integrate the trend data sequence Td, the indicator sequence Ki and the suggestion sequence As to obtain the risk assessment report R.

[0113] This embodiment achieves a comprehensive risk assessment and grading of disconnection faults through a systematic risk assessment and analysis strategy. First, by analyzing the subsystem characteristics and change trends, the risk value of each subsystem is accurately calculated. Secondly, by analyzing the risk propagation coefficient and the coupling relationship between subsystems, the risk level of the overall system can be accurately assessed. Thirdly, through risk threshold analysis and grading, a complete risk rating system is established. Finally, by generating a detailed risk assessment report, comprehensive risk control recommendations are provided. This embodiment not only provides accurate risk warnings, but also provides a reliable basis for system maintenance decisions, effectively improving the operational safety and reliability of the system. Through the monitoring and dynamic updating of key risk indicators, potential failure risks can be discovered in a timely manner, and corresponding preventive measures can be recommended.

[0114] According to one aspect of the present application, a method for detecting a control loop disconnection is specifically as follows: a controller detects whether a control loop disconnection detection function is enabled. If not enabled, disconnection detection is not performed. If the control loop disconnection detection function is enabled, it is determined whether the control relay in the control loop is closed. If the control relay is closed, disconnection detection is not performed. If the control relay is disconnected, the controller detects whether there is a low-level input. If there is a low-level input, the control loop is not disconnected. If no low-level signal input is detected, the control loop is determined to be disconnected. Due to the limitation of the current in the control loop, the traditional solution of directly detecting the loop state through an optocoupler is not feasible. Compared with the traditional solution of detecting loop disconnection, the circuit of this embodiment adds a filtering circuit and a signal conditioning circuit. Through reasonable configuration parameters, the weak signal is converted into a level signal that can be recognized by the controller, thereby realizing the completion of the control loop disconnection detection under low current conditions.

[0115] According to one aspect of the present application, a control loop disconnection detection circuit includes: a current limiting circuit, an isolation circuit, a filtering circuit, a signal conditioning circuit and a controller. The current limiting circuit, the isolation circuit, the filtering circuit and the signal conditioning circuit are connected in series in sequence, and the output end of the signal conditioning circuit is connected to the signal input end of the controller. The current limiting circuit is used to limit the current in the disconnection detection circuit to the microampere level. The isolation circuit is used to isolate the strong current in the control loop from the weak current of the controller. The filtering circuit is used to filter out some interference signals in the signal loop. The signal conditioning circuit is used to convert the weak electric signal into a level pulse signal that can be recognized by the controller and transmit the signal to the controller. The controller is used to detect whether a periodic level pulse signal or a low-level signal is received. If so, it is judged that the disconnection detection circuit is normal. If a high-level signal is always received, it is judged that the disconnection detection circuit is disconnected. The current limiting circuit uses a current limiting resistor, and multiple resistors are connected in series. The isolation circuit uses a bidirectional optocoupler isolation, which can achieve complete electrical isolation, and can realize AC and DC detection, and has no polarity requirements for DC voltage. The filtering circuit uses capacitor filtering.

[0116] When a disconnection detection circuit of a control loop of this embodiment is used, the disconnection detection circuit is connected in parallel in a control loop composed of external control COM1, OPE1, and a control relay, and the controller is used to send a control signal of the relay action to the control relay. When the controller controls the control relay to disconnect, the disconnection detection circuit forms a series loop with the power supply and circuit breaker tripping in the control loop. The current is limited by the current limiting circuit and then isolated by the optocoupler circuit. The optocoupler secondary sends the weak signal to the signal conditioning circuit after filtering. The signal conditioning circuit converts the small signal into a level signal that can be recognized by the controller, and then the controller detects the level of this signal to determine whether the line is disconnected. When the control loop is disconnected, the detection point always maintains a high level, which can be judged as a disconnection condition; when there is AC power in the control loop, the level of the detection point will change with the change of the control voltage. When the control voltage is high, the detection point is low level. When the control voltage becomes low, the detection point will appear high level, and the high level pulse width will become wider as the control voltage becomes lower, until it is always high and cannot be recognized. When there is DC power in the control loop, the level of the detection point will become low level.

[0117] The present invention realizes intelligent management of the entire process from signal acquisition to predictive maintenance by constructing a complete control loop disconnection detection solution. First, in the signal preprocessing stage, the high quality of the basic data is ensured through adaptive sampling frequency optimization and multi-dimensional signal quality evaluation; secondly, the accurate capture of disconnection features is achieved through feature extraction and selection mechanisms; thirdly, the accuracy of fault diagnosis is improved through multi-level fault identification and credibility evaluation; finally, predictive maintenance is achieved through timing prediction and risk assessment. By combining traditional control theory with modern artificial intelligence technology, the present invention can adaptively adjust its parameters and strategies to continuously improve the accuracy and reliability of detection. It not only improves the detection and prediction capabilities of disconnection faults, but also reduces maintenance costs, improves the reliability and safety of the system, and provides a strong guarantee for the continuity and stability of industrial production.

[0118] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for detecting disconnection in a control loop, characterized in that: The steps include: S1. Obtain an original signal data set, optimize the sampling frequency of the original signal data set to obtain an optimized sampling frequency value; based on the optimized sampling frequency value, resample the original signal data set to obtain a resampled signal set; perform noise processing on the resampled signal set to obtain a preprocessed signal set; Perform data quality assessment on the preprocessed signal set and generate quality assessment indicators; The original signal data set includes current signal, voltage signal, temperature signal and vibration signal; S2. Based on the preprocessed signal set and the quality assessment index, signal enhancement processing is performed to obtain an enhanced signal set; Extract time domain and frequency domain features from the enhanced signal set to obtain a feature matrix; Perform correlation analysis on the feature matrix to obtain the optimized feature set; Decompose the optimized feature set according to time series to obtain a time series feature vector; Perform stability assessment on time series feature vectors and generate reliable feature sets and reliability reports; S3, based on the reliable feature set and the reliability report, perform feature space mapping to obtain a mapping feature set; perform hierarchical feature extraction on the mapping feature set to obtain a hierarchical feature set; Based on the hierarchical feature set, fault identification is performed to obtain the fault probability distribution; the credibility of the fault probability distribution is evaluated to obtain the credibility index; based on the credibility index and the fault probability distribution, anomaly identification is performed to obtain the anomaly assessment report and the final diagnosis result; S4, extracting time series features based on the abnormality assessment report, the final diagnosis result and the pre-stored historical diagnosis database to obtain a time series pattern set; Based on the time series pattern set, trend prediction is performed to obtain a prediction result sequence; based on the prediction result sequence, risk assessment is performed to obtain a risk assessment report; Based on the risk assessment report and the prediction result sequence, a decision plan is generated to obtain an optimized decision plan.

2. A control loop disconnection detection method according to claim 1, characterized in that: Step S1 is further as follows: S11, obtaining an original signal data set, calculating the statistical features of each signal within a preset time window, and obtaining a feature vector; The values ​​in the feature vector are normalized to obtain a standardized feature vector; a weight coefficient group is determined based on the degree of numerical deviation in the standardized feature vector; the weight coefficient group is multiplied by the standardized feature vector to obtain a weighted feature vector; S12, based on the preset sampling frequency range, calculate the change rate of the weighted feature vector to obtain the change rate value; calculate the sampling frequency adjustment coefficient according to the change rate value and the current sampling frequency; multiply the sampling frequency adjustment coefficient by the current sampling frequency to obtain a new sampling frequency; determine whether the new sampling frequency is within the sampling frequency range, if within the range, directly use the new sampling frequency as the optimized sampling frequency value; if out of the range, take the boundary value and output the optimized sampling frequency value; S13, based on the optimized sampling frequency value, interpolating the original signal data set to obtain an interpolated data set; performing anti-aliasing filtering on the interpolated data set to obtain a filtered data set; performing timed sampling on the filtered data set according to the optimized sampling frequency value to obtain a resampled signal set; S14, obtaining a resampled signal set, calculating a local fluctuation feature of the signal, and obtaining a fluctuation feature set; Based on the fluctuation feature set, filter parameters are determined to obtain a filter parameter set; Adaptively filter the resampled signal set using the filter parameter set to obtain a preprocessed signal set; S15, calculating the signal-to-noise ratio of the signal based on the preprocessed signal set to obtain a signal-to-noise ratio value group; Based on the preprocessed signal set, the continuity index of the signal is calculated to obtain a continuity index group; based on the preprocessed signal set, the integrity index of the signal is calculated to obtain an integrity index group; The signal-to-noise ratio value group, continuity index group and integrity index group are normalized and weighted to obtain the quality assessment index.

3. A control loop disconnection detection method according to claim 2, characterized in that: Step S2 is further as follows: S21, based on the quality assessment index, calculating the enhancement coefficient to obtain an enhancement coefficient group; performing a time domain enhancement operation on the enhancement coefficient group and the preprocessed signal set to obtain time domain enhancement data; performing a frequency domain enhancement operation on the enhancement coefficient group and the preprocessed signal set to obtain frequency domain enhancement data; Fusing the time domain enhanced data and the frequency domain enhanced data to obtain an enhanced signal set; S22. Based on the enhanced signal set, calculate the first-order difference of the signal to obtain a first difference sequence; Based on the first difference sequence, the second-order difference of the signal is calculated to obtain a second difference sequence; Extracting time domain features according to the first difference sequence and the second difference sequence to obtain a time domain feature sub-matrix; performing spectrum analysis on the enhanced signal set to obtain a frequency domain feature sub-matrix; merging the time domain feature sub-matrix and the frequency domain feature sub-matrix to obtain a feature matrix; S23, based on the feature matrix, calculating the correlation coefficients between the columns of the matrix to obtain a correlation coefficient matrix; According to the correlation coefficient matrix, the importance score of each feature is calculated to obtain the importance score vector; According to the importance score vector, feature selection is performed on the feature matrix to obtain the optimized feature set; S24, based on the optimized feature set, segmenting the feature data according to a preset fixed time window to obtain a time series group; extracting a trend component from the time series group to obtain a trend feature vector; Extract the periodic component from the time series group to obtain the periodic feature vector; extract the mutation component from the time series group to obtain the mutation feature vector; combine the trend feature vector, the periodic feature vector and the mutation feature vector to obtain the time series feature vector; S25. Based on the time series feature vector and the preprocessed signal set, calculate the time stability index of the feature to obtain a stability index group; based on the time series feature vector and the preprocessed signal set, calculate the anti-interference ability index of the feature to obtain an anti-interference index group; filter the features according to the stability index group and the anti-interference index group to obtain a reliable feature set; generate a reliability report based on the evaluation data of the screening process.

4. A control loop disconnection detection method according to claim 3, characterized in that: Step S3 is further as follows: S31. Based on the reliability report, calculate the feature weights to obtain a weight matrix; perform mapping transformation on the weight matrix and the reliable feature set to obtain an initial mapping set; perform dimensional optimization on the initial mapping set to obtain a mapping feature set; S32, extracting mutation feature parameters based on the mapping feature set to obtain a mutation feature subset; Based on the mapping feature set, the fluctuation feature parameters are extracted to obtain the fluctuation feature subset; Based on the mapping feature set, trend feature parameters are extracted to obtain a trend feature subset; Combine the mutation feature subset, the fluctuation feature subset and the trend feature subset to obtain a hierarchical feature set; S33, based on the hierarchical feature set, calculating the time domain identification index to obtain a time domain index group; based on the hierarchical feature set, calculating the frequency domain identification index to obtain a frequency domain index group; based on the hierarchical feature set, calculating the power domain identification index to obtain a power index group; performing probability conversion on the time domain index group, the frequency domain index group and the power index group to obtain a fault probability distribution; S34, based on the fault probability distribution and the pre-stored historical accuracy data, calculate the evidence strength index to obtain the evidence strength value; obtain the current working condition data, calculate the working condition similarity, and obtain the similarity value; calculate the comprehensive credibility according to the evidence strength value and the similarity value to obtain the credibility index; S35. Based on the credibility index, fault probability distribution and pre-configured normal operating condition model, calculate the deviation between the current state and the normal model to obtain the deviation index; calculate the degree of abnormality according to the deviation index and the credibility index to obtain the abnormality assessment report; determine the final diagnosis result based on the abnormality assessment report and the fault probability distribution.

5. A control loop disconnection detection method according to claim 4, characterized in that: Step S4 is further as follows: S41, based on the abnormal assessment report, the final diagnosis result and the pre-stored historical diagnosis database, extract the historical diagnosis sequence features and the current diagnosis features to obtain the historical feature sequence and the current feature sequence; calculate the time series correlation between the historical feature sequence and the current feature sequence to obtain the correlation matrix; Based on the correlation matrix, the key evolution features are extracted to obtain the temporal pattern set; S42, based on the time series pattern set, construct a state prediction model to obtain a prediction model parameter group; according to the prediction model parameter group, calculate the future state sequence to obtain a prediction state sequence; calculate the confidence interval of the prediction state sequence to obtain an interval boundary sequence; combine the prediction state sequence and the interval boundary sequence to obtain a prediction result sequence; S43, based on the prediction result sequence, calculating the risk value of each subsystem to obtain a subsystem risk group; According to the subsystem risk groups, calculate the overall system risk and obtain the system risk value; Based on the system risk value, determine the current risk level and obtain the risk level matrix; Integrate the system risk value and risk level matrix to obtain a risk assessment report; S44, generating a set of optional decision solutions based on the risk assessment report and the prediction result sequence to obtain a decision solution set; Calculate the safety score of each solution in the decision solution set to obtain a safety score group; calculate the cost score of each solution in the decision solution set to obtain a cost score group; The timeliness score of each plan in the decision plan set is calculated to obtain a timeliness score group; based on a comprehensive evaluation of the safety score group, cost score group and timeliness score group, an optimized decision plan is obtained.

6. A control loop disconnection detection method according to claim 5, characterized in that: Step S11 is further as follows: S111, obtaining an original signal data set, performing block processing on each signal within a preset time window to obtain a signal block sequence; calculating the mean of each signal block sequence to obtain a mean sequence; Calculate the standard deviation of each signal block sequence to obtain a standard deviation sequence; calculate the peak value of each signal block sequence to obtain a peak value sequence; combine the mean value sequence, standard deviation sequence and peak value sequence to obtain a single signal feature vector; S112, obtaining the current signal and the voltage signal in the original signal data set, calculating the phase difference sequence, and obtaining the phase feature vector; obtaining the temperature signal in the original signal data set, and calculating the thermal response feature based on the temperature signal and the current signal to obtain the thermal feature vector; obtaining the vibration signal in the original signal data set, and calculating the coupling feature based on the vibration signal and the voltage signal to obtain the coupling feature vector; combining the phase feature vector, the thermal feature vector, and the coupling feature vector to obtain the combined feature vector; S113, based on the single signal feature vector, calculating the rate of change of the feature to obtain a rate of change sequence; based on the combined feature vector, calculating the trend of the feature to obtain a trend sequence; based on the rate of change sequence and the trend sequence, constructing a time series matrix to obtain a time series feature vector; S114, based on the single signal feature vector, the combined feature vector and the time series feature vector, calculate the covariance matrix of each feature vector to obtain a correlation matrix; based on the correlation matrix, perform feature combination to obtain a feature vector; The eigenvector is normalized to obtain a standardized eigenvector; the discrete degree of each component in the standardized eigenvector is calculated to obtain a weight coefficient group; the weight coefficient group is multiplied by the standardized eigenvector to obtain a weighted eigenvector.

7. A control loop disconnection detection method according to claim 5, characterized in that: Step S13 is further as follows: S131, based on the original signal data set and the optimized sampling frequency value, calculate the sampling point spacing to obtain a spacing sequence; based on the spacing sequence, perform segmentation processing to obtain a segment sequence; based on the segment sequence, calculate the local features of each segment of data to obtain a feature sequence; Select the optimal interpolation algorithm according to the feature sequence to obtain the interpolation data set; S132, based on the interpolation data set, calculating the signal spectrum distribution to obtain a spectrum sequence; based on the spectrum sequence, constructing filter parameters to obtain a filter parameter group; applying the filter parameter group to perform bandpass filtering to obtain a bandpass signal sequence; performing phase correction on the bandpass signal sequence to obtain a filtered data set; S133. Obtain a filtered data set, calculate the sampling moment according to the optimized sampling frequency value, and obtain a sampling moment sequence; calculate the signal value of each sampling moment sequence to obtain a sampling value sequence; based on the sampling value sequence, perform a sampling consistency check to obtain a consistency index sequence; perform sampling correction according to the consistency index sequence to obtain a resampled signal set.

8. A control loop disconnection detection method according to claim 5, characterized in that: Step S22 is further as follows: S221, based on the enhanced signal set, calculating the local extreme value points of the signal to obtain an extreme value sequence; based on the extreme value sequence, calculating the signal envelope to obtain a first envelope sequence; calculating the first-order difference of the enhanced signal set to obtain a first difference sequence; Calculate the second-order difference of the first difference sequence to obtain the second difference sequence; Calculating the statistical characteristics of the first difference sequence and the second difference sequence to obtain a time domain statistical sequence; combining the first envelope sequence and the time domain statistical sequence to obtain a basic time domain feature vector; S222, based on the enhanced signal set, performing short-time Fourier transform to obtain a time-frequency spectrum matrix; performing energy distribution analysis on the time-frequency spectrum matrix to obtain an energy distribution sequence; calculating the frequency center drift of the time-frequency spectrum matrix to obtain a center drift sequence; combining the energy distribution sequence and the center drift sequence to obtain a frequency domain basic feature vector; S223, based on the basic time domain feature vector and the frequency domain basic feature vector, calculate the phase relationship between the features to obtain a phase relationship matrix; based on the basic time domain feature vector and the frequency domain basic feature vector, calculate the time-frequency coupling degree of the features to obtain a coupling degree sequence; based on the phase relationship matrix and the coupling degree sequence, construct a joint feature to obtain a time-frequency joint feature vector; S224. Based on the basic time domain feature vector, grouping is performed according to feature correlation to obtain a time domain feature sub-matrix; based on the frequency domain basic feature vector and the time-frequency joint feature vector, the main feature components are extracted to obtain a frequency domain feature sub-matrix; feature standardization is performed on the time domain feature sub-matrix and the frequency domain feature sub-matrix to obtain a standardized feature group; the standardized feature group is reorganized according to a preset format to obtain a feature matrix.

9. A control loop disconnection detection method according to claim 5, characterized in that: Step S32 is further as follows: S321, based on the mapping feature set, calculating the instantaneous change rate of the signal to obtain a change rate sequence; performing threshold segmentation on the change rate sequence to obtain a segmented sequence; based on the segmented sequence, calculating the jump amplitude of each segment of the signal to obtain an amplitude sequence; based on the segmented sequence, calculating the jump time interval between adjacent segments to obtain a time series; combining the amplitude sequence and the time series to obtain a mutation feature subset; S322, based on the mapping feature set, calculating the local fluctuation period to obtain a periodic sequence; based on the periodic sequence, calculating the fluctuation envelope to obtain a second envelope sequence; Performing frequency decomposition on the second envelope sequence to obtain a frequency sequence; Based on the second envelope sequence, the energy distribution of the fluctuation is calculated to obtain an energy sequence; The frequency sequence and energy sequence are combined to obtain a subset of fluctuation characteristics; S323, based on the mapping feature set, applying adaptive polynomial fitting to obtain a fitting coefficient sequence; Based on the fitting coefficient sequence, the fitting residual sequence is calculated; Based on the fitting residual sequence, the trend slope is extracted to obtain the slope sequence; based on the slope sequence, the acceleration characteristics of the trend are calculated to obtain the acceleration sequence; the slope sequence and the acceleration sequence are combined to obtain the trend feature subset; S324, based on the mutation feature subset, the fluctuation feature subset and the trend feature subset, calculating the correlation between the feature subsets to obtain a correlation matrix; Based on the correlation matrix, the feature importance is sorted to obtain the importance sequence; according to the importance sequence, the features are screened and reorganized to obtain a reorganized feature matrix; the reorganized feature matrix is ​​organized into a hierarchical structure to obtain a hierarchical feature set.

10. A control loop disconnection detection method according to claim 5, characterized in that: Step S42 is further as follows: S421, based on the time series pattern set, calculating the autocorrelation coefficient of the time series data to obtain a correlation coefficient sequence; based on the correlation coefficient sequence, determining the order of the prediction model to obtain the order parameter; according to the order parameter, extracting the key state variables to obtain a state variable sequence; normalizing the state variable sequence to obtain a prediction model parameter group; S422, based on the prediction model parameter group and the most recent state variable sequence, calculate the short-term prediction sequence; calculate the error range of the short-term prediction sequence to obtain a short-term error sequence; perform drift correction on the short-term prediction sequence to obtain a corrected sequence; combine the corrected sequence and the short-term error sequence to obtain a short-term prediction result; S423, calculating the mid-term trend characteristics based on the prediction model parameter group and the pre-stored historical trend data to obtain a trend characteristic sequence; based on Trend feature sequence, perform progressive forecasting, and obtain a mid-term forecast value sequence; Calculate the uncertainty of the mid-term forecast value sequence to obtain the uncertainty sequence; combine the mid-term forecast value sequence and the uncertainty sequence to obtain the mid-term forecast result; S424, based on the short-term prediction results and the mid-term prediction results, calculate the weight coefficient of each prediction result to obtain a weight sequence; The short-term prediction results and the medium-term prediction results are fused according to the weight sequence to obtain the prediction state sequence; based on the prediction state sequence, the confidence level of the fused prediction is calculated to obtain the interval boundary sequence; the prediction state sequence and the interval boundary sequence are combined in chronological order to obtain the prediction result sequence.

Images