Intelligent equipment remote maintenance autonomous monitoring method

Abstract

The invention discloses an intelligent equipment remote maintenance autonomous monitoring method, and relates to the technical field of intelligent equipment operation and maintenance, and the method comprises the following steps: building a unified time baseline, deploying a zero delay bypass sampling mechanism, extracting an unsmooth energy transition track from continuously collected multi-source signals, and generating an event sequence of full time domain resolution; a transition confidence window set is constructed based on an event sequence, a pulse signal is analyzed by using a high-order cumulant, and key signal fragments are screened in combination with a sparse peak value extraction method. According to the method, a closed-loop regulation and control system driven by a unified time baseline is constructed, event sequence capture, high-order statistical analysis and time reversal intervention are fused, real-time recognition and active hedging of equipment transition energy are achieved, the fault prediction and self-healing capacity is improved, the service life of the structure is prolonged, and the operation and maintenance cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment operation and maintenance technology, specifically to a method for remote maintenance and autonomous monitoring of intelligent equipment. Background Technology

[0002] Remote maintenance and autonomous monitoring of intelligent equipment refers to the real-time collection and transmission of the operational status of intelligent devices distributed in different locations to a remote monitoring center through communication networks, sensing technology, and artificial intelligence algorithms. The system autonomously completes the entire process of status identification, fault diagnosis, maintenance demand prediction, and alarm notification without frequent manual intervention. Its core lies in two aspects: firstly, utilizing multi-source sensor data and equipment operation logs to monitor equipment around the clock, promptly identifying potential hazards and performance degradation trends; secondly, through built-in intelligent decision-making models and adaptive strategies, automatically generating maintenance instructions or remote control operations, thereby achieving predictive maintenance and self-healing management of the equipment. This method not only improves the intelligence and automation level of equipment operation and maintenance, reduces the cost of manual inspections and reactive repairs, but also significantly improves the safety, stability, and lifespan of the equipment.

[0003] The existing technology has the following shortcomings: In existing technologies, monitoring equipment operation status typically relies on dynamic smoothing algorithms to filter and reduce noise in collected multi-source signals, mitigating the impact of environmental disturbances and instantaneous fluctuations on monitoring results. However, during high-speed equipment operation, energy transition signals with extremely short durations but significant amplitudes often appear. These signals often contain potential risks such as component fatigue accumulation and abnormal local loads. Due to the over-reliance on dynamic smoothing algorithms in existing technologies, such energy transitions are often masked as random noise, failing to trigger timely warning mechanisms. This results in imperceptible cumulative impact effects during subsequent operation, ultimately easily leading to sudden fractures of critical structures, causing serious safety hazards and economic losses.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a remote maintenance and autonomous monitoring method for intelligent devices to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a remote maintenance and autonomous monitoring method for intelligent devices, comprising the following steps: Establish a unified time baseline, deploy a zero-delay bypass sampling mechanism, extract unsmoothed energy transition trajectories from continuously acquired multi-source signals, and generate event sequences with full time-domain resolution; Based on the event sequence, pulse signals are analyzed using high-order cumulants, and key signal segments are screened by combining sparse peak extraction methods to construct a set of transition confidence windows. Under the constraints of the transition confidence window set, anti-smoothing processing is performed and pulse details are traced back by reversible time-frequency transformation to reconstruct the transition energy curve; Dynamic response modeling is performed based on the transition energy curve, the coupling relationship between energy input and local load of the equipment is calculated, the fatigue residual energy distribution is extracted, key transition sections are identified, and a causal proof chain is established. A feedforward discriminator is constructed based on a causal proof chain, and a phase-driven probe is superimposed to generate an early warning threshold and determine the instantaneous response cycle. Based on the instantaneous processing cycle start-up time inversion control, the programmable energy injection device is driven to apply phase conjugate micro-excitation in the zero-phase neighborhood, establish a reverse energy channel, suppress transition energy, and update the dynamic calibration factor to a unified time baseline, forming a closed-loop control process.

[0007] The preferred steps for generating event sequences with full time-domain resolution are as follows: A unified time baseline is established, a high-precision time synchronization device is used to provide the master clock pulse signal, each signal acquisition channel is initialized and phase calibration is performed to ensure that the sampled data of each channel has absolute time consistency. A zero-delay bypass sampling process is deployed on the basis of a unified time baseline. A high-speed analog-to-digital converter is used to directly extract the original physical signal from the dual output and record a unified timestamp, thus preserving complete transient characteristics. Based on the original sampled data, energy transition segment extraction processing is performed. The energy change rate and change acceleration are obtained by differential method. The effective transition position is determined by combining the background energy statistics and the corresponding signal segment is extracted. An event sequence is constructed according to the order in which the transition segments occur. Cooperative transition conditions are set to filter out pseudo-events. The events are organized into a structured sequence containing fields such as time, amplitude, and background features, forming an event representation with full temporal resolution.

[0008] Preferably, the steps for constructing the transition confidence window set are as follows: Based on the event sequence, the start time, end time, maximum amplitude change, duration, and adjacent background signal features of each event are extracted to construct a candidate signal segment with unified time alignment. Perform high-order cumulant calculations on candidate signal segments, extract skewness, kurtosis and peak frequency, determine whether there are non-Gaussian pulse abrupt changes, and construct a set of abrupt change segments; Sparse peak extraction is performed on the mutation fragments to extract key peak structures that meet the requirements of amplitude variation, rise time duration and asymmetry characteristics, and to screen out waveform fragments with diagnostic significance. A set of transition confidence windows is constructed based on key waveform segments, a structured confidence index is added and a confidence score is calculated, and the effective transition windows are determined for subsequent analysis.

[0009] Preferably, the steps for reconstructing the transition energy curve are as follows: Under the constraint of the transition confidence window set, the corresponding segment of the confidence window is extracted from the original unfiltered signal, the anti-smoothing operator is performed to restore the weakened pulse structure, and a dynamic amplitude tracking factor is introduced to correct the high steepness waveform. Multi-level wavelet decomposition is performed on the signal segment after anti-smoothing, energy change segments are identified and target point sets are constructed, inverse wavelet transform is performed based on the target point set and boundary signals are fused to complete the time-frequency reconstruction of pulse details; Based on the amplitude change and energy shift characteristics between the reconstructed signal and the signal before antismoothing, a recovery scaling factor is generated and energy amplitude calibration is performed using a sliding correction method, ultimately outputting a complete transition energy curve.

[0010] Preferably, the process of establishing the causal proof chain is as follows: Based on the transition energy curve, a dynamic response modeling relationship is established for each key area of ​​the equipment structure. The coupling mapping between energy input and local load of the structure is calculated point by point, and the abnormal response state caused by high-frequency pulse excitation is identified. Based on the stress response history of structural nodes and material fatigue limit parameters over a continuous time period, the fatigue residual energy distribution is extracted to identify the key transition segment entering the fatigue acceleration stage. A causal analysis was conducted on the stress response process of the critical transition section and the corresponding transition energy input to construct a causal proof chain with temporal logic and variable correlation between transition input, local response and fatigue degradation.

[0011] Preferably, in the process of constructing the causal proof chain, four indicators are used to verify the causal relationship between energy input and structural response: linear correlation coefficient, temporal coherence, average time lag and response slope change. The transition event is determined to be a valid causal node if the scores of three or more indicators exceed the set threshold.

[0012] Preferably, the steps for determining the instantaneous processing rhythm are as follows: A feedforward discriminator is constructed based on the key transition segment of the causal proof chain output, extracting the peak transition energy, the abrupt change point of structural stress gradient and the inflection point of fatigue residual energy decrease, and merging them into a three-dimensional state sequence on a unified time baseline. After the feedforward discriminator predicts the high-risk area, an early phase micro-amplitude perturbation signal is applied to the corresponding structural area, and the response waveform is collected for comparison of spectrum and time domain characteristics to verify the structural softening trend. Based on the verification results, a structural status scoring index system was established to calculate the risk level and generate early warning thresholds. By combining the start point of the transition event with the structural response delay, an instantaneous processing time window is determined for subsequent intervention and control operations.

[0013] Preferably, based on the instantaneous processing cycle start-up time inversion control, the energy injection device is driven to apply phase conjugate micro-excitation in the zero-phase neighborhood, establishes a reverse energy channel to suppress transition energy, and updates the dynamic calibration factor to the unified time baseline. The steps are as follows: The start time of the energy inversion operation is determined based on the instantaneous processing cycle, and an inversion excitation signal that is symmetrical in the time domain, has the same frequency and amplitude, and is opposite in phase to the target transition signal is constructed. Under a unified time baseline, an excitation gating time interval is set, and the inversion signal is applied to the target structure region through a high-speed signal interface within this time interval. The inversion excitation signal is converted into an analog electrical signal and applied to the excitation interface of the target structure, thereby forming an energy counteracting effect in the structural medium that is opposite in phase to the original transition signal. After excitation is completed, the response waveform of the target structure is acquired, the frequency, phase and amplitude changes before and after excitation are compared, the response state transition matrix is ​​constructed and a dynamic calibration factor is generated. The dynamic calibration factor is written into the unified time baseline to correct the excitation parameters and intervention strategies for subsequent transition events, thus constructing a traceable closed-loop control process.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention constructs a full-process state perception and adaptive control system with a unified time baseline as the time sequence main line, high-fidelity event sequence as the information core, causal proof chain as the discrimination criterion, and closed-loop energy intervention as the response mechanism. This system not only achieves high-sensitivity capture of key transition energy events at microsecond-level time resolution, but also, by combining high-order statistical analysis and anti-smoothing signal reconstruction technology, fully preserves the original transient pulse characteristics, thereby constructing a high-fidelity transition energy curve reflecting the true load state of the structure. This supports quantitative modeling and dynamic evaluation of structural fatigue evolution, impact response mechanisms, and fault trends. In particular, by introducing a control strategy based on time inversion theory and an energy injection mechanism based on phase conjugate micro-excitation, real-time counter-intervention of abnormal energy and on-site repair and adjustment of the structural response are achieved, fundamentally promoting the transformation of equipment from a passive alarm mode to an active self-healing mechanism. This method comprehensively improves the intelligence, foresight, and response accuracy of equipment operation status monitoring, significantly enhances the service life and operational safety of key components under complex load environments, and reduces maintenance frequency and operating costs. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the method for remote maintenance and autonomous monitoring of intelligent devices according to the present invention. Detailed Implementation

[0017] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0018] This invention provides, for example Figure 1 The remote maintenance and autonomous monitoring method for intelligent devices shown includes the following steps: Establish a unified time baseline, deploy a zero-delay bypass sampling mechanism on this unified time baseline, extract the complete energy transition trajectory without smoothing from continuously acquired multi-source signals, and generate an event sequence with full time domain resolution; By constructing a unified time baseline and combining it with a zero-delay bypass sampling process, high-fidelity extraction and complete reconstruction of energy transition features in multi-source signals are achieved, thereby forming an event sequence with full time-domain resolution. The specific steps include: Establishing a unified time baseline. To ensure a strictly consistent time reference for continuous physical signals from different sources, this step employs a high-precision time synchronization device equipped with a temperature-controlled crystal oscillator, configured as the sole time standard for all signal acquisition paths. During implementation, each independent acquisition channel is first initialized using the master clock pulse signal output from the synchronizer, ensuring its sampling trigger logic maintains strict edge alignment with the master clock signal. Next, a highly stable phase-locked loop (PLL) circuit is configured in each acquisition channel to capture and calibrate the master clock signal in real time, eliminating minor time offsets caused by differences in cable length and internal processing delays. Based on this, the timestamp of each signal sample is uniformly assigned by the master synchronization clock, achieving absolute time consistency across all physical acquisition sources. To verify time consistency, a full-channel timestamp readback operation is performed every second. If a time deviation exceeding 1 microsecond is detected in any channel, an automatic synchronization correction process is triggered, ensuring the entire acquisition network maintains high-precision time consistency during long-term operation. The unified time baseline established in this way provides a verifiable and traceable absolute time reference for all subsequent signal analysis, which is significantly better than traditional methods based on frame rate or logical time windows.

[0019] A zero-delay bypass sampling process is deployed based on a unified time baseline. The purpose of this step is to fully preserve the transient details of all original physical signals, avoiding the masking of high-frequency content by filtering or noise reduction processing in the main path. In implementation, each signal acquisition point is equipped with dual output ports: one for the regular processing path, connected to subsequent smoothing and compression processes; and the other directly connected to a high-speed analog-to-digital converter, forming a parallel bypass sampling channel. The analog-to-digital converter used in the bypass channel has a response time of less than 10 nanoseconds and a sampling rate exceeding 500 MSPS, enabling high-fidelity digitization of physical signals without preprocessing. All sampled signals are written to a local buffer in real time and simultaneously marked with a unified timestamp, ensuring no sorting errors or time drift occur during data transmission and processing. Under this structure, all unfiltered, uncompressed, and unresampled original signals are preserved, providing a clean and undistorted physical basis for subsequent analysis, significantly improving the ability to capture minute transitions, and is particularly suitable for identifying high-frequency transient impacts, small energy mutations caused by fatigue, or high-order fluctuations caused by abnormal mechanical contact.

[0020] Using the acquired raw high-fidelity signals, precise extraction of energy transition segments is performed. In this stage, each signal is first aligned and interpolated according to a unified time baseline, ensuring each channel has corresponding sample values ​​at every time point, constructing a complete cross-signal time matrix. Subsequently, first and second-order differential operations are performed on each channel in chronological order to extract the energy change rate and acceleration between consecutive sampling points. If the energy change rate between two consecutive sampling points exceeds a set threshold (e.g., a 50% relative amplitude increase) and the duration is less than 2 milliseconds, this position is preliminarily identified as the transition start point. To ensure accuracy, the system extracts 2 milliseconds of background signal segments before and after each transition start point, calculating their average energy level and variance. If the energy difference between the transition segment and the preceding and following background exceeds three standard deviations, it is confirmed as a valid transition segment. This process is executed in parallel for all signal channels. Finally, the selected transition segments from each channel are summarized, located at their absolute time points, and their start time, end time, amplitude range, and preceding and following background trend characteristics are recorded. This method differs significantly from traditional methods that rely on global averages or empirical judgments. It can dynamically adapt to the energy fluctuation baseline of different devices and operating states, thereby improving the sensitivity and resistance to false judgments in transition identification.

[0021] Based on the identified transition segments, a full-time-domain event sequence is constructed to form a unified representation structure. In this step, each transition segment is arranged in chronological order of its occurrence, generating a one-dimensional time-stream sequence. Each transition event is treated as an independent object, containing the following fields: event start time, event end time, signal source, energy change rate, energy peak, duration, and the corresponding differences in background energy statistics before and after the event. To avoid spurious mutation interference, each transition event must satisfy the condition that at least two dissimilar signal sources exhibit coordinated energy mutations in adjacent time periods before it can be included in the final event sequence. In terms of data structure organization, the sequence is stored using a time-indexed method and supports fast retrieval according to any time interval, ensuring efficient response capabilities in subsequent modeling processing. In addition, all event segments undergo time alignment and resampling before storage to ensure that events from different sources can be used for causal analysis, trend modeling, and intervention decisions on a unified time axis. The final output event sequence possesses high temporal accuracy, high physical realism, and strong contextual relevance, laying a solid signal foundation for subsequent multi-level state judgment and fatigue degradation trend prediction.

[0022] Based on the event sequence, higher-order cumulants are used to analyze the pulse signal components in the event sequence. Combined with the sparse peak extraction method, key signal segments are screened, and a set of transition confidence windows is constructed. After constructing the full-time-domain energy transition event sequence, in order to identify abnormal pulse signals with diagnostic value and eliminate non-causal perturbations, it is necessary to perform in-depth statistical and waveform analysis on the event sequence and construct a set of transition confidence windows with time accuracy and confidence criteria. The specific steps include: Based on the generated energy transition event sequence, key parameters for each event are extracted for initial screening analysis. For each event, its absolute start time, end time, maximum amplitude change, duration, and the average energy change trend of the 500 adjacent sampling points before and after the event are extracted. Subsequently, based on the start and end times, a time window is formed by extending forward by 1 millisecond and backward by 1 millisecond, respectively, to form a time window with a total length of 2 milliseconds plus the original event duration. High-fidelity signal segments are extracted within this window to form preliminary candidate segments. To ensure the consistency of subsequent analysis results, during the extraction process, the candidate segments of each channel are interpolated to a time resolution of one sampling point per microsecond using a unified time baseline, ensuring that all signals are processed on the same time frame. Peak amplitude normalization is also used to unify the amplitude of different channels. This step not only provides a unified input signal structure for subsequent higher-order statistical analysis but also avoids feature recognition distortion caused by time sampling errors or channel amplitude differences.

[0023] Higher-order statistical features are extracted from the constructed candidate signal segments to identify non-stationary pulse signal components. Within each candidate segment, several overlapping sliding time windows are defined, each with a length of 200 microseconds and a sliding step size of 20 microseconds. Statistical calculations are performed segment by segment. For the signal within each sliding window, the skewness (third-order cumulant), kurtosis (fourth-order cumulant), and peak frequency (sixth-order cumulant) are calculated to reflect the signal's asymmetry, steepness, and extreme abrupt changes within that interval. After statistical analysis, the obtained higher-order feature values ​​are compared with the statistical baseline values ​​of the background signal 1 millisecond before and after the corresponding candidate segment. If the skewness or kurtosis offset from the baseline is greater than 20%, and the peak frequency is higher than the reference threshold, the time window is initially marked as a potential pulse activity interval. Subsequently, the start and end times of all potential intervals and their corresponding signal channel numbers are recorded to construct a set of higher-order abrupt change segments, which serves as the input basis for the next step of sparse peak extraction. Compared with traditional judgment logic based on time-domain energy threshold, this method can more accurately detect real non-Gaussian pulse components caused by physical structural changes, energy release from material microcracks, bearing impacts, etc.

[0024] Based on the results of higher-order statistical features, a sparse peak extraction operation is performed within the aforementioned abrupt change segments to extract the most representative key waveform structures. In this process, the abrupt change segments are first resampled to a fixed resolution at 1-microsecond intervals, and a candidate list of local peaks is constructed. This candidate list contains all peaks and troughs in each waveform, along with their corresponding time points, slopes before and after, rise time, fall time, local curvature, and waveform asymmetry indices. Subsequently, background fluctuations are eliminated through rule-based filtering: if the rise time of a peak is less than 0.3 milliseconds, the peak amplitude is more than 50% higher than the average of the previous 20 microseconds, and the fall time asymmetry is greater than 1.5 times, then this peak is identified as a valid transition point. Furthermore, multiple consecutive valid transition points are grouped into peak events, and their center position, pulse width, energy surge ratio, and tail energy decay characteristics are calculated. Based on these, the waveform is segmented, retaining key signal segments with diagnostic significance. The waveform constructed using this method not only retains the burst characteristics of the original signal in its temporal structure but also highlights its anomalous components in the energy dimension, providing a precise foundation for subsequent modeling and backtracking.

[0025] Based on the extracted key waveform segments, a complete set of transition confidence windows is constructed, and a structured confidence index is attached to each window. Each transition confidence window consists of a start time, end time, signal source channel, peak amplitude, duration, higher-order statistical eigenvalues, sparse peak positions, and contextual change indicators, and is indexed by event number, corresponding one-to-one with the aforementioned event sequence. To further improve the reliability of the identification results, a comprehensive scoring mechanism is introduced during the construction of the confidence window set. A confidence score is calculated for each window, which consists of three parts: higher-order statistical offset weight, key peak structural integrity score, and inter-channel temporal coherence. The scoring standard is a percentage system. When the confidence score is higher than 75, the window is determined to be a valid transition window; if the score is between 60 and 75, it is considered a window to be confirmed; and those below 60 are directly removed from the analysis set. The final set of transition confidence windows can be used to support subsequent anti-smoothing signal reconstruction, energy evolution path extraction, and fatigue life prediction modeling. Its completeness and accuracy are significantly better than existing wave signal processing methods that rely on time window slicing or single index judgment, demonstrating the technical leadership and engineering adaptability of this method in weak pulse signal identification.

[0026] Under the constraints of the transition confidence window set, the anti-smoothing operator is executed, and the pulse signal details are traced back point by point in combination with the reversible time-frequency transformation to reconstruct the transient pulses that were masked by the smoothing process, and the transition energy curve calibrated by the energy amplitude is output. After constructing the set of transition confidence windows, in order to recover the transient pulse information suppressed by traditional smoothing and improve signal integrity and energy characterization capabilities, a point-by-point reconstruction operation needs to be performed on the original signal under the constraint of the confidence window. The specific steps include: Based on the start and end times provided by the transition confidence window set, anti-smoothing processing is performed on each high-confidence abrupt signal segment to restore the original waveform structure. In practice, continuous signal segments are extracted from the original unfiltered signal according to the time intervals marked in the confidence window as the target processing segment. For the target processing segment, the specific smoothing method used in traditional signal processing is first evaluated, such as whether it is a weighted moving average or whether there is low-pass filtering within a fixed period, and the mathematical transformation relationship corresponding to this smoothing process is constructed. On this basis, the sampled signal is back-calculated to obtain the theoretical pulse profile of the signal segment before filtering. To more realistically restore the details of signal changes, a multi-level iterative process is set up in this back-calculation process. After each iteration, the difference between the recovered pulse and the original signal is corrected to ensure that no miscompensation is caused by high-frequency noise. For high-steep pulses that are significantly weakened during smoothing, a "dynamic amplitude tracking factor" is introduced in the recovery phase. By extracting stable mean and variance from the background segments before and after the pulse, and combining this with the rising slope of the pulse itself, the original maximum amplitude range is inferred, thus completing the boundary stretching of the spike signal. This method is more robust than traditional filter deconvolution methods, avoiding phase mismatch problems caused by incomplete frequency response, and is particularly suitable for use in low-speed rotating machinery and operating environments where periodic pulse interference is prone to aliasing.

[0027] For the signal segment obtained after anti-smoothing processing, a reversible time-frequency transform with time positioning capability is used to backtrack and reconstruct the pulse details point by point. Operationally, wavelet basis functions (such as bioorthogonal wavelets or antisymmetric Daubechies wavelets) with asymmetry, tight support, and strong local analytical capabilities are first selected to perform multi-level discrete wavelet decomposition on the anti-smoothed signal. Each wavelet decomposition layer decomposes the signal into two components: approximate signals and detail signals, with the detail components mainly retaining high-frequency transition information. At each scale, the local energy concentration of the detail signal and the energy jump amplitude between adjacent sampling points are calculated. If the energy surge amplitude within a certain time interval is greater than three times the average level of that scale, and the abrupt change segment length is less than 0.5 milliseconds, it is marked as the region where the true pulse component is located. All time-domain locations satisfying the above conditions are cross-matched across all scales. If there is a high energy coupling phenomenon at the same time point in the high-frequency and mid-frequency scales, the time point is considered to have multi-scale pulse consensus and is thus included in the backtracking reconstruction target point set. Subsequently, an inverse wavelet transform was performed on the target point set to recover the reconstructed signal segment containing all the details of the real pulses. To improve the signal splicing accuracy, a linear fusion region with window function weights was set in the transition region between the recovered signal and the original anti-smoothed signal to ensure smooth waveforms and continuous spectra at boundary positions, without abrupt changes or energy jumps. After reconstruction, a continuous Fourier transform was performed on the spectrum of the entire signal segment to verify that its frequency distribution did not contain any pseudo-high-frequency components, ensuring the spectral reliability of subsequent analyses.

[0028] After completing the time-frequency backtracking reconstruction, to ensure that the reconstructed pulse signal accurately reflects the impact intensity of the real physical process in terms of amplitude and energy, it is necessary to perform energy amplitude calibration processing based on multi-index fusion. In practice, firstly, the peak amplitude, waveform envelope area, rise time, fall time, duration, and energy difference between each reconstructed pulse and the adjacent background signal within the local time window are calculated. Then, the amplitude change rate and energy offset of the reconstructed waveform before and after anti-smoothing are compared. Combined with the previously constructed anti-smoothing process modeling results, the degree of signal compression of each pulse is derived, and the corresponding recovery scaling factor is generated. When performing amplitude calibration, a three-step strategy of "difference mapping - measured back substitution - sliding correction" is adopted: first, the local peak is initially corrected using the recovery scaling factor; then, the corrected amplitude is interpolated to fill the entire pulse envelope range; finally, the waveform slope and curvature curves are fitted by a sliding window to achieve continuous adjustment of the entire pulse signal profile. The calibrated pulse segments are then compared again with the original mutation indices marked in the confidence window. If the deviation is controlled within ±10%, the recovery result is considered to meet the actual engineering requirements. All signal segments processed in the above manner are combined according to the time sequence and channel sequence to finally output a complete, coherent transition energy curve with real physical meaning. This curve not only retains the time position, amplitude information, and structural details of various pulse signals, but also reflects the real input intensity of each transition event on the equipment load and structural stress at the amplitude level, providing a reliable energy-driven basis for subsequent dynamic response modeling.

[0029] Based on the transition energy curve, a dynamic response modeling process is performed to calculate the coupling relationship between energy input and local load of the equipment point by point, extract the fatigue residual energy distribution, identify the key transition segment formed over time, and establish a causal proof chain. To deeply identify the actual impact of transition energy on the equipment structure, dynamic response modeling is carried out based on the reconstructed transition energy curve. The mapping relationship between energy input and structural response is established point by point, the fatigue residual energy distribution is extracted, and key transition segments in structural evolution are identified, thereby constructing a causal proof chain based on time logic. Specifically, the following steps are included: Based on the reconstructed and amplitude-calibrated transition energy curves, dynamic response modeling is performed on various key areas of the equipment structure, and the mapping relationship between pulse energy input and local structural loads is calculated point by point. In the specific implementation, the transition energy curves are first divided into multiple non-overlapping segments in chronological order, with each segment length set to 0.5 seconds for local response modeling. For the transition energy input within each time period, its effective area is defined, and an energy-stress coupling model is established by combining the elastic modulus of the actual structural material, the moment of inertia of the local structure, the stiffness coefficient, and the support boundary conditions. This model is used to estimate the maximum stress amplitude, local strain gradient, and elastic potential energy distribution induced by each energy pulse in the target area. All calculations are based on the conservation of Lagrange kinetic and potential energy, and point-by-point interpolation is used to convert the energy input into a structural response curve. Subsequently, nodal stress comparison is performed on each key structural component, and structural nodes that repeatedly exceed the average level by more than 30% within a short period of time are considered to be in an abnormal response state under high-frequency pulse excitation. This method differs from traditional approaches that rely on empirical formulas or simplified models based on static loads. By introducing time-varying energy pulse inputs combined with the local dynamic characteristics of the structure, it improves the timeliness, specificity, and structural coupling accuracy of the modeling.

[0030] Based on the established dynamic response relationship, the distribution characteristics of residual fatigue energy are gradually extracted during continuous structural loading to identify critical transition sections that may form in local areas during long-term operation. In the implementation process, the stress response history of each structural node over multiple consecutive time periods is first unfolded temporally to form a multidimensional time-stress-energy matrix. Then, relying on the fatigue theory of metal structures, the historical energy input received by each node is compared and analyzed with the allowable fatigue limit parameters of its material. The percentage of remaining load-bearing capacity for each node is calculated using the fatigue life margin calculation method. For example, if the low-cycle fatigue limit of a node's material is below 3 MPa per millimeter and its cumulative life is 10^6 cycles, and the calculation results show that its cumulative received energy has reached 85% of the allowable total, then the node is considered to have entered the critical region of residual fatigue energy. Furthermore, considering the changes in the stress growth rate within the three most recent transition stages, if the growth rate gradually accelerates and differs from the previous growth stages, the node is preliminarily determined to have entered the fatigue acceleration zone. All nodes within the acceleration zone are arranged chronologically to identify the key locations and time periods on the structural fatigue evolution path where the energy mutation tendency first occurs, defined as "critical transition sections." These transition sections are not only precursors to changes in structural performance, but also priority areas for proactive intervention or real-time early warning. Unlike common methods that estimate fatigue life using global static cumulative loads, this invention introduces a local dynamic energy inflow mechanism and stress response relationship, enabling a more precise reveal of the actual path of fatigue degradation at the microscale.

[0031] After identifying the critical transition segment of structural fatigue evolution, a causal analysis is performed on the energy input, structural response evolution process, and fatigue deterioration segment in the transition energy curve to gradually establish a causal proof chain with clear temporal logic and strong variable correlation. The specific method is as follows: For all transition pulses before each "critical transition segment", extract their corresponding time point, energy amplitude, load concentration area, and stress response intensity to establish an event index. Then, starting from each structural response change point, retrieve all transition energy events that may affect the response and establish a preliminary event flow in chronological order. Based on this, the following four causal relationship verification indicators are set: (1) the linear correlation coefficient between energy input and the maximum local stress; (2) the temporal synergy between the local elastic strain of structural nodes and the frequency of transition pulses; (3) the average time lag between the rate of decrease of fatigue residual energy and the cumulative input of continuous pulses; (4) the correlation between the degree of convergence of transition pulses and the slope of the change in structural response speed. The above indicators are scored sequentially. If three or more scores exceed a set threshold (e.g., a correlation coefficient greater than 0.85), the transition event can be identified as a valid causal node leading to structural degradation. Finally, these causal events are linked together in a time-chain manner to form a complete causal proof chain of "transition input—local response—fatigue degradation—critical transition segment." Each proof chain includes specific time sequences, structural locations, energy triggering sources, stress response indicators, and residual fatigue assessment values. It possesses strong readability and engineering traceability, and can be directly used as input for subsequent early warning threshold determination and response action initiation logic.

[0032] Based on the causal proof chain, a feedforward discriminator is constructed, and a micro-amplitude phase traction probe is superimposed to quantify and verify the strength of the key transition segment, generate an early warning threshold, and determine the instantaneous response rhythm for dynamic intervention. Based on constructing a causal proof chain and identifying key transition segments, to achieve early prediction and precise intervention in risk evolution regions, it is necessary to further construct a feedforward discriminator, combined with micro-amplitude phase traction probing, to achieve physical empirical verification of the structural state, ultimately forming an early warning threshold and defining the intervention rhythm. The entire process includes the following steps: A feedforward discriminator is constructed based on the established causal proof chain to predict the evolution trend of critical states. The specific steps include: extracting three types of time series data from each causal proof chain: pulse peaks and frequencies in the transition energy curve, stress gradient abrupt change points in the structural response curve, and decreasing inflection points in the fatigue residual energy curve. These time series are fused along a unified time baseline to form a three-dimensional time coordinate space, within which a multi-dimensional state assessment model is constructed. The assessment model is based on a combination logic of physical indicators: if, within two consecutive time periods, the transition energy input continuously increases by more than 30%, the response rate of the corresponding structural node increases synchronously by more than 25%, and the slope of the fatigue residual energy decreases increases and the trend continues for more than 5 seconds, then this region is marked as a high-risk prediction segment. Based on this, the feedforward discriminator uses a sliding time window method to re-evaluate the score of each structural node in the risk prediction model every 200 milliseconds, and outputs the prediction priority ranking according to the score. Compared to traditional methods that set trigger thresholds based on a single abnormal indicator, this method uses multidimensional physical state evolution trends for discrimination, avoiding misjudgments caused by short-term fluctuations or occasional events, and significantly improving the stability and advance capability of prediction and discrimination.

[0033] After the feedforward discriminator outputs a high-risk region, a micro-amplitude phase traction test is performed to enhance the physical verifiability of the state judgment results, thereby observing the actual response change trend of the structure. This test is implemented as follows: The structurally sensitive area corresponding to the high-risk prediction segment is selected as the intervention target, such as stress-concentrated welds on continuous plate and shell structures, interference fit parts of bearings in rotating machinery, and abrupt stiffness changes at frame connections. A micro-amplitude perturbation excitation signal with an amplitude not exceeding 5% of the original response amplitude, consistent frequency, but phase advanced by 90 degrees is applied using a phase synchronization exciter. The duration of the signal application does not exceed 15 milliseconds to ensure that the normal operating state of the structure is not disrupted. Immediately after signal injection, synchronous acquisition of structural response data begins, analyzing amplitude changes, response delay time, and spectral center drift in the waveform. If, under the perturbation, the target structure's response amplitude increases by more than 10% compared to before the perturbation, the spectral center frequency shifts to lower frequencies by more than 5 Hz, and there is a significant phase lag, then the structure is confirmed to be in a critical softening state, indicating that the high-risk judgment given by the feedforward discriminator has physical verification support. Compared with traditional passive response observation methods, this type of response testing method based on phase traction probing is proactive, controllable, and has high resolution, and can effectively identify early signs of structural degradation that have not yet become apparent.

[0034] Based on the state prediction results of the feedforward discriminator and the physical response verification results of the micro-amplitude phase traction test, a quantitative analysis of the intensity of the key transition section is performed, and a multi-level linkage early warning threshold is generated. In this step, a comprehensive structural state scoring index system is constructed, with scoring dimensions including: energy input change rate, structural response slope, fatigue remaining energy depletion rate, and test response sensitivity. Each index is normalized to a score from 0 to 1 and assigned a weight ratio (e.g., 0.3, 0.3, 0.2, 0.2), and a total risk score is synthesized. Risk levels are classified according to the score: 0 to 0.4 is low risk, 0.4 to 0.6 is medium risk, 0.6 to 0.8 is high risk, and above 0.8 is extremely high risk. In high-risk and extremely high-risk states, an early warning signal is immediately generated, and the target structural location, trigger time, and suggested response action are written into the warning information list. To avoid false alarms or missed alarms, a lag mechanism is introduced when the score approaches the boundary; the risk level is upgraded only if the upward trend of the score continues for three cycles and the direction is stable. This method differs from traditional strategies that rely on a fixed single threshold for judgment. Instead, it is based on the real-time structural state evolution trend, which improves the accuracy of early warning and its dynamic adaptability.

[0035] After confirming the structural risk status, to achieve a rapid and effective intervention response, it is necessary to determine the instantaneous response timeframe for intervention based on the time location of the transition event and the response delay characteristics. The implementation method is as follows: using the start time of the transition event at each high-risk structural point as a reference time, the time interval from 2 milliseconds before to 5 milliseconds after that time point is defined as the target intervention interval. Combining the actual delay characteristics of the response signal, the point of maximum response speed change within the target interval is further selected as the peak timeframe. Boundary times of 0.5 milliseconds are set before and after the peak timeframe to form the final response timeframe window, with a total window duration not exceeding 2 milliseconds. Intervention measures must be triggered within this window, such as through micro-amplitude reverse-phase impacts, vibration cancellation excitation, and rapid adjustment of structural interface pressure, to achieve real-time suppression of the response process. The process of determining the response timeframe considers energy transfer paths, structural reaction speed, and control delay limitations, possessing practical execution value and differing from traditional response methods that rely on preset periodic control. By precisely locating the disposal window, we ensure that the control actions have minimal coupling error and maximum energy suppression efficiency, providing a rigorous timing basis for the next step of execution time-reversal gating and energy extinction.

[0036] Based on the instantaneous processing cycle, the time-reversal gating control process is initiated, driving the programmable energy injection device to apply phase conjugate micro-excitation in the zero-phase neighborhood, establishing a reverse energy channel, realizing real-time suppression of transition energy, and updating the dynamic calibration factor to a unified time baseline, forming a closed-loop control process from signal acquisition to energy suppression. After clarifying the instantaneous response timeline for high-risk transition events, a time-reversal gating control process needs to be constructed to achieve immediate response and proactive intervention to pulse energy. This involves establishing a reverse energy conduction path through phase-conjugated micro-excitation and dynamically updating the time baseline based on the excitation response results, thereby achieving complete closed-loop control from event identification to energy suppression. The specific steps of this process are as follows: Based on the identified instantaneous processing beat, the start time of the energy inversion operation is determined, and an inversion excitation signal that is completely symmetrical to the target pulse transition waveform is constructed. In this step, the complete pulse trajectory of the target transition signal in the time domain is extracted, and this trajectory is reversed and arranged at the original sampling accuracy to generate an inversion signal sequence with equal amplitude and period but opposite phase. The inversion signal is synchronized with the original signal in time, but a 180-degree mirror flip operation is performed in phase to ensure that it can completely cancel out the original transition signal when applied. While generating the inversion signal, the dominant frequency component, harmonic composition, and energy envelope shape are extracted according to the frequency characteristics of the original transition signal to ensure that the inversion signal has an equivalent spectral distribution in the frequency domain and avoid the suppression ineffectiveness problem caused by frequency mismatch. Finally, the inversion signal is encapsulated into an excitation command with fixed length, fixed amplitude, and fixed phase, waiting for the gating window to release it.

[0037] Energy inversion gating is initiated within a precise time window, releasing the phase-conjugate micro-excitation signal through a high-time-resolution gating mechanism. This step synchronously retrieves the start timestamp of the processing cycle from a unified time baseline, constructing a gating activation segment of 2 to 5 milliseconds to ensure that the excitation signal does not overlap with other signals during application. During gating, the inverted signal is transmitted in real time to the excitation transmitter via a high-speed digital signal output interface, and its output waveform is controlled to be transmitted within the zero-phase neighborhood, ensuring complete overlap with the target pulse signal on the time axis, achieving time-domain cancellation. Compared to traditional preset excitation methods, this method offers higher time synchronization accuracy and response speed, avoiding intervention failures caused by delays. It is particularly suitable for equipment structures with high-frequency, short-duration energy bursts, such as high-speed rotors, piezoelectric components, or micro-actuators.

[0038] An energy emission device with precise signal control is driven to apply inversion excitation to the target structural region, achieving physical energy suppression. Specifically, the main release path of the target transition pulse energy is first identified, and an excitation point is selected along this path, such as the cantilever root of a metal structure, the fastening connection surface between components, or the boundary between a panel and a reinforcing rib. The conjugate excitation signal is converted into a high-fidelity analog electrical signal and then applied to the excitation interface (such as a piezoelectric ceramic plate, electromagnetic actuator, or thermoelectric coil) in the target region via a power amplifier. This generates a micro-excitation waveform on the target structure with the same amplitude, frequency, and phase opposite to the original transition signal. This excitation waveform travels in reverse along the original energy propagation path after entering the structural medium, forming an energy counter-current zone with the releasing original transition energy, reducing stress concentration and fatigue damage to the structure. Immediately after the excitation process ends, the response waveform of the target area structure is acquired for comparison and analysis. If the relative energy attenuation rate reaches a set threshold (e.g., above 50%), energy suppression is considered successful.

[0039] After the energy reverse intervention is completed, the response results of the target structure are retrospectively evaluated, and a dynamic calibration factor is generated to adjust the identification and excitation parameters of subsequent transition events. This step uses synchronous measurement to collect response signals from the same structural parts before and after excitation, comparing their peak frequency, phase delay, amplitude changes, and energy envelope area changes. After normalizing the various indicators, a three-dimensional response change model is constructed. This model generates a structural response state transition matrix to determine whether stress path changes, wave drag changes, or elastic modulus drift have occurred. If a trend is confirmed, a dynamic calibration data package containing time-series labels, location labels, and correction factors is generated, using structural coordinate points as indices and the change amplitude as calibration values. This calibration factor is written into a unified time baseline in real time to automatically match the corresponding excitation parameters, timing, and intervention strategies when the next transition event arrives, achieving dynamic evolution and self-optimization of the control logic.

[0040] All operational processes are constructed as a closed-loop control chain, forming a complete closed-loop structural response control system from energy sensing, cycle decision-making, signal inversion, conjugate excitation, suppression verification to feedback calibration. In the final stage, all excitation operations and response results are synchronously recorded in the energy control process trajectory, constructing a continuous control response chain in chronological order. Each link in the chain is marked with the event trigger time, intervention response time, excitation waveform parameters, structural feedback indicators, and calibration factor values, and a retrospective comparison is performed before the next intervention to determine whether the control strategy needs parameter migration or logic adjustment. In this way, the originally isolated event control operations are transformed into a dynamic control network that can accumulate, learn, and evolve, continuously optimizing the energy management strategy during long-term operation, and ultimately achieving the adaptive fatigue resistance evolution of the equipment structure under multi-cycle fatigue loads.

[0041] This invention constructs a full-process state perception and adaptive control system with a unified time baseline as the time sequence main line, high-fidelity event sequence as the information core, causal proof chain as the discrimination criterion, and closed-loop energy intervention as the response mechanism. This system not only achieves high-sensitivity capture of key transition energy events at microsecond-level time resolution, but also, by combining high-order statistical analysis and anti-smoothing signal reconstruction technology, fully preserves the original transient pulse characteristics, thereby constructing a high-fidelity transition energy curve reflecting the true load state of the structure. This supports quantitative modeling and dynamic evaluation of structural fatigue evolution, impact response mechanisms, and fault trends. In particular, by introducing a control strategy based on time inversion theory and an energy injection mechanism based on phase conjugate micro-excitation, real-time counter-intervention of abnormal energy and on-site repair and adjustment of the structural response are achieved, fundamentally promoting the transformation of equipment from a passive alarm mode to an active self-healing mechanism. This method comprehensively improves the intelligence, foresight, and response accuracy of equipment operation status monitoring, significantly enhances the service life and operational safety of key components under complex load environments, and reduces maintenance frequency and operating costs.

[0042] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for autonomous monitoring of remote maintenance of intelligent equipment, characterized in that, The method comprises the following steps: Establish a unified time baseline, deploy a zero-delay bypass sampling mechanism, extract unsmoothed energy transition trajectories from continuously collected multi-source signals, and generate event sequences with full time-domain resolution; Based on the event sequence, use high-order cumulants to analyze the pulse signal, combine the sparse peak extraction method to filter key signal segments, and construct a transition confidence window set; Under the constraint of the transition confidence window set, perform anti-smoothing processing and combine the reversible time-frequency transform to backtrack the pulse details, and reconstruct the transition energy curve; Based on the transition energy curve, model the dynamic response, calculate the coupling relationship between energy input and device local load, extract the fatigue residual energy distribution, identify the key transition section, and establish a causal proof chain; Based on the causal proof chain, construct a feedforward discriminator, superimpose a phase traction probe, generate an early warning threshold, and determine the instantaneous disposal tempo; Based on the instantaneous disposal tempo, start time inversion regulation, drive the programmable energy injection device to apply a phase conjugate micro-excitation in the zero-phase neighborhood, establish a reverse energy channel, suppress the transition energy, and update the dynamic calibration factor to the unified time baseline. 2.The intelligent device remote maintenance self-monitoring method of claim 1, wherein, The event sequence generation step with full time-domain resolution is as follows: A unified time baseline is constructed, a high-precision time synchronization device is used to provide a master clock pulse signal, each signal acquisition channel is initialized and phase calibration is performed, so that the sampling data of each channel has absolute time consistency; On the basis of the unified time baseline, a zero-delay bypass sampling process is deployed, an analog-to-digital conversion device with high-speed response capability is used to directly extract the original physical signal from the dual-channel output and record the unified time stamp, and the complete transient characteristics are retained; Based on the original sampling data, perform energy transition segment extraction processing, use the difference method to obtain the energy change rate and acceleration, and combine the background energy statistics to determine the effective transition position and extract the corresponding signal segment; According to the order of transition segment occurrence, an event sequence is constructed, a cooperative transition condition is set to exclude false events, events are organized into a structured sequence, and an event expression with full time-domain resolution is formed. 3.The intelligent device remote maintenance self-monitoring method of claim 2, wherein, The transition confidence window set construction steps are as follows: Based on the event sequence, the start time, end time, maximum amplitude change, duration and adjacent background signal characteristics of each event are extracted, and a unified time-aligned candidate signal segment is constructed; High-order cumulant calculation is performed on the candidate signal segment, skewness, kurtosis and peak frequency are extracted, it is judged whether there is a non-Gaussian pulse mutation, and a mutation segment set is constructed; In the mutation segment, sparse peak extraction processing is performed, key peak structures that meet the amplitude change, rising edge duration and non-symmetry characteristics requirements are extracted, and waveform segments with diagnostic significance are selected; According to the key waveform segment, a transition confidence window set is constructed, a structured confidence index is added and a confidence score is calculated, and an effective transition window is determined for subsequent analysis. 4.The intelligent device remote maintenance self-monitoring method of claim 3, wherein, The transition energy curve reconstruction steps are as follows: Under the constraint of the transition confidence window set, the confidence window corresponding section is extracted from the original unfiltered signal, the anti-smoothing operator processing is performed, the weakened pulse structure is restored, and a dynamic amplitude tracking factor is introduced to correct the high-steepness waveform; The signal segment after the inverse smoothing processing is subjected to multi-level wavelet decomposition, the energy mutation section is identified, and a target point set is constructed, inverse wavelet transformation is performed according to the target point set, and boundary signals are fused to complete the time-frequency reconstruction of the pulse details; According to the amplitude change and energy offset characteristics between the reconstructed signal and the signal before inverse smoothing, a recovery scale factor is generated, and a sliding correction method is used to perform energy amplitude calibration, and finally a complete transition energy curve is output. 5.The intelligent device remote maintenance self-monitoring method of claim 4, wherein, The causal proof chain establishment process is as follows: Based on the transition energy curve, a dynamic response modeling relationship is established for each key area of the device structure, the coupling mapping between energy input and local load of the structure is calculated point by point, and the abnormal response state caused by high-frequency pulse excitation is identified; Based on the stress response history of the structure node in the continuous time period and the material fatigue limit parameter, the fatigue residual energy distribution is extracted, and the key transition section entering the fatigue acceleration stage is identified; The stress response process of the key transition section and the corresponding transition energy input are causally analyzed to construct a causal proof chain with time logic and variable correlation between transition input, local response and fatigue deterioration. 6.The intelligent device remote maintenance self-monitoring method of claim 5, wherein, In the process of establishing the causal proof chain, four indicators, linear correlation coefficient, time domain cooperation degree, average time lag and response slope change, are used to verify the causal relationship between energy input and structure response, and more than three indicators are scored above the set threshold to determine that the transition event is an effective causal node. 7.The intelligent device remote maintenance and repair autonomous monitoring method according to claim 5, characterized in that, The instantaneous disposal tempo determination steps are as follows: Based on the key transition section output by the causal proof chain, a feedforward discriminator is constructed, the transition energy peak, the structure stress gradient mutation point and the fatigue residual energy inflection point are extracted, and they are fused into a three-dimensional state sequence on a unified time baseline; After the high-risk area is predicted by the feedforward discriminator, a micro-disturbance signal with an advanced phase is applied to the corresponding structure area, and the response waveform is collected for frequency spectrum and time domain feature comparison to verify the structure softening trend; According to the verification result, a structure state scoring index system is established, the risk level is calculated, and an early warning threshold is generated; Combined with the starting point of the transition event and the structure response delay, the instantaneous disposal tempo window is determined, which is used for subsequent execution of intervention control operation. 8.The intelligent device remote maintenance self-monitoring method of claim 7, wherein, Based on the instantaneous disposal tempo start time inversion regulation, the energy injection device is driven to apply a phase conjugate micro-excitation in the zero-phase neighborhood to establish a reverse energy channel to suppress the transition energy, and the dynamic calibration factor is updated to the unified time baseline steps are as follows: Based on the instantaneous disposal tempo, the energy inversion operation start time point is determined, and an inversion excitation signal which is symmetrical in time domain, equal in frequency amplitude and opposite in phase with the target transition signal is constructed; Set the excitation gate time section under the unified time baseline, and control the inversion signal to be applied to the target structure area through the high-speed signal interface within the time section; The inversion excitation signal is converted into an analog electric signal and loaded into the excitation interface of the target structure to form an energy hedging effect opposite in phase to the original transition signal in the structure medium; After the excitation is completed, the target structure response waveform is collected, the frequency, phase and amplitude changes before and after the excitation are compared, the response state transition matrix is constructed, and the dynamic calibration factor is generated; The dynamic calibration factor is written into a unified time baseline, used to correct the excitation parameters and intervention strategies of subsequent transition events, and a traceable closed-loop control process is constructed.

Images