Intelligent fault diagnosis method and system for intelligent pod

Abstract

The present application provides an intelligent fault diagnosis method and system for an intelligent pod. Among them, the method synchronously collects the power line voltage fluctuation signal and the motor control line transient signal, and performs multi-band feature extraction and transient energy density analysis respectively. Based on the synchronous trigger point, a comparative analysis mechanism of the voltage fluctuation amplitude and the transient amplitude is established, the time difference is calculated, and a composite rule containing the amplitude threshold interval and the phase correlation value is generated. By constructing a multi-dimensional spatial model that integrates amplitude, energy, and time difference, a fault prediction rule library is established, and finally the abnormal point distribution pattern is identified through the feature vector matching degree, the fault type is accurately located and the diagnosis result is output, thereby realizing the fault diagnosis and positioning of the intelligent pod. The technical solution provided by the present application realizes the accurate fault positioning and diagnosis of the intelligent pod power supply and control system through multi-dimensional signal feature fusion and intelligent matching algorithm.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent diagnosis of electromechanical equipment of intelligent pods, and in particular to an intelligent fault diagnosis method and system for intelligent pods. Background Art

[0002] In equipment such as drone inspections and ship propulsion systems, smart pods serve as core sensors and control units, and their operating status directly impacts mission reliability. These pods must achieve millisecond-level coordinated control of multiple subsystems, including power systems, motor drives, and communication modules, in complex electromagnetic environments and under severe vibration conditions. This places systematic demands on fault diagnosis technology, requiring the construction of a full-link monitoring system covering multi-dimensional features such as power fluctuations, transient shocks, and signal distortion to ensure comprehensive awareness of the operating status of each subsystem. Diagnostic algorithms must be developed that dynamically adapt to different operating conditions, meet the timeliness requirements for abnormal response, and complete closed-loop processing from signal acquisition and feature extraction to fault location in milliseconds. Furthermore, a self-learning fault knowledge base must be established to leverage accumulated fault cases and expert experience to enable autonomous identification of non-standard fault modes and predictive maintenance, reducing the frequency of human intervention.

[0003] The current mainstream solution for this requirement is a pod fault diagnosis system based on fault tree analysis. This system constructs a logically gated fault tree model, breaking down top-level events like steering failure and communication interruption into lower-level events like inverter anomalies and sensor failures. This system then uses a historical fault database for probabilistic reasoning. By establishing a hierarchical fault causality model, this approach effectively analyzes the propagation paths of known fault modes, providing a theoretical basis for maintenance decisions. It also incorporates multi-sensor fusion technology to enhance data collection.

[0004] The shortcomings of existing solutions are concentrated in three aspects. First, the reasoning mechanism based on the historical fault library is difficult to deal with new complex faults, such as cascading failures caused by the superposition of power supply harmonic interference and control signal jitter. The lack of cross-domain correlation analysis capabilities leads to insufficient recognition rate of unknown faults. Second, the fusion analysis of multi-source heterogeneous data is insufficient. The existing system only implements single-dimensional feature extraction and lacks cross-domain correlation modeling capabilities, resulting in the omission of key fault characteristics. Finally, the diagnostic algorithm mostly uses an offline training mode and cannot dynamically adjust threshold parameters according to the actual operating conditions of the pod. The diagnostic accuracy decreases significantly when the dynamic environment changes, seriously restricting the ability to continue operating under complex conditions. These limitations have created significant bottlenecks in dynamic environment adaptability, multi-source data collaborative analysis, and knowledge iteration efficiency. It is urgent to overcome the shortcomings of existing technologies through the development of a new diagnostic framework. Summary of the Invention

[0005] The present application provides an intelligent fault diagnosis method and system for an intelligent pod, which is used to solve the problems in the prior art of insufficient multi-dimensional feature correlation analysis, poor adaptability to dynamic environments, and low recognition rate of complex faults.

[0006] In a first aspect, the present application provides an intelligent fault diagnosis method for an intelligent pod, comprising:

[0007] Obtain voltage fluctuation signals of the smart pod power line and transient signals of the motor control line;

[0008] Performing multi-band processing on the voltage fluctuation signal to generate voltage segment features, and recording the amplitude parameters of the voltage fluctuation signal; performing feature extraction on the transient signal to generate a transient peak by capturing energy density changes in the signal amplitude mutation region, and recording the energy parameters of the transient signal;

[0009] Based on the voltage segment characteristics and the transient peak value, a signal trigger point synchronization mechanism is established, and based on the synchronization trigger point, a comparative analysis of the voltage fluctuation amplitude and the transient amplitude is performed to calculate the time difference between the two;

[0010] Based on a preset periodic reference signal, in combination with the amplitude parameter and the energy parameter, a composite rule including an amplitude threshold interval and a phase correlation value is generated;

[0011] Constructing a multidimensional space including the amplitude parameter, the energy parameter, and the time difference, and establishing a fault type prediction model in the multidimensional space based on the composite rule;

[0012] By calculating the matching degree between the distribution position of the feature vector in the multi-dimensional space and the fault type prediction model, the abnormal points are identified, and the corresponding fault type is determined according to the distribution pattern of the abnormal points in the multi-dimensional space, and the diagnostic results including the identified fault type and the corresponding abnormal points are output.

[0013] Optionally, the method of calculating the degree of match between the distribution position of the feature vector in the multidimensional space and the fault type prediction model to identify abnormal points, determining the corresponding fault type based on the distribution pattern of the abnormal points in the multidimensional space, and outputting a diagnosis result including the identified fault type and the corresponding abnormal point includes:

[0014] Converting the feature vector into a coordinate point in the multidimensional space through a preset nonlinear coordinate mapping function, calculating the geometric distance between the coordinate point and the historical benchmark data form in the fault type prediction model, and marking the corresponding coordinate point whose geometric distance exceeds a dynamic threshold as an abnormal point;

[0015] Applying a dual filtering mechanism including spatial filtering and temporal filtering to the outliers to screen out outliers that meet both spatial aggregation and temporal continuity characteristics to generate an outlier set;

[0016] Input the abnormal point set into the regional feature analysis module, extract the distribution geometric characteristics and time series characteristics of the abnormal point set, map the fault type label according to the distribution geometric characteristics and time series characteristics, and assign the corresponding abnormal point;

[0017] The fault type label and the timestamp and spatial coordinates of the corresponding abnormal point are encapsulated into a diagnostic data packet that complies with the industrial Internet of Things protocol to output the diagnostic result.

[0018] Optionally, the step of inputting the abnormal point set into a regional feature analysis module, extracting distribution geometric features and time series features of the abnormal point set, mapping fault type labels according to the distribution geometric features and time series features, and assigning corresponding abnormal points includes:

[0019] In the regional feature analysis module, the coordinate span of the outlier point set in three-dimensional space is counted to determine the spatial distribution range, and the average distance between adjacent points is calculated to evaluate the spatial dispersion of the outlier point set. The spatial distribution range and spatial dispersion are used as distribution geometric features;

[0020] At the same time, the frequency distribution of the outliers in the outlier set within a unit time is counted, and a time interval sequence is generated according to the difference between the timestamps of adjacent outliers. The frequency distribution and the time interval sequence are combined as time series features;

[0021] The spatial distribution feature and the time series feature are fused to form a composite feature descriptor, and the composite feature descriptor is matched with the historical fault samples in the fault type prediction model for similarity. The fault type label corresponding to the matched fault mode is assigned to the corresponding abnormal point through feature similarity calculation.

[0022] Optionally, constructing a multidimensional space including the amplitude parameter, the energy parameter, and the time difference, and establishing a fault type prediction model in the multidimensional space based on the composite rule includes:

[0023] Arranging the amplitude parameter, the energy parameter, and the time difference in chronological order into three-dimensional data units containing a timestamp identifier, and constructing a multi-dimensional space;

[0024] Based on the composite rule and in combination with geometric distribution characteristics of the three-dimensional data units in the multi-dimensional space, clustering regions satisfying the composite rule are divided in the multi-dimensional space;

[0025] Performing statistical analysis on the clustering area, extracting the cluster center coordinates, cluster range boundary values, and density distribution characteristics of the coordinate points within the cluster to form cluster feature data;

[0026] The cluster feature data is matched with typical fault features in a pre-stored fault type library to generate a fault type prediction model including cluster center coordinate weights, dynamic adjustment coefficients of cluster range boundaries, and density distribution matching thresholds.

[0027] Optionally, establishing a signal trigger point synchronization mechanism based on the voltage segment characteristics and the transient peak value, and performing a comparative analysis of the voltage fluctuation amplitude and the transient amplitude based on the synchronization trigger point to calculate the time difference between the two includes:

[0028] generating a dual constraint condition based on the amplitude duration and the amplitude change slope, screening the voltage abnormal signal segment whose amplitude parameter exceeds a first threshold in the voltage segment feature based on the dual constraint condition, and locating the energy mutation region whose energy parameter mutation exceeds a second threshold in the transient peak based on the consistency of timestamp alignment and energy density gradient direction;

[0029] Establish a signal trigger point synchronization mechanism to compare the time overlap of the voltage abnormal signal segment and the energy mutation region point by point, and select matching events with timestamp differences less than a preset tolerance to mark them as synchronization trigger points;

[0030] Taking the synchronous trigger point as a reference, the amplitude change slope of the voltage fluctuation signal before and after the trigger is calculated, and the energy density change rate of the transient peak before and after the trigger is calculated at the same time. The time difference between the voltage fluctuation amplitude and the starting moment of the transient amplitude is determined according to the intersection of the amplitude change slope and the energy density change rate.

[0031] Optionally, the generating of a composite rule including an amplitude threshold interval and a phase correlation value based on a preset periodic reference signal in combination with the amplitude parameter and the energy parameter includes:

[0032] Synchronously dividing a preset periodic reference signal into a plurality of periodic units of equal length according to the power frequency period of the power line;

[0033] In each of the periodic units, an allowable fluctuation range of the amplitude parameter is set based on the statistical distribution of historical normal operating condition data to generate an amplitude threshold interval;

[0034] Determining a phase correlation value between the energy parameter and the start time of the periodic unit according to the distribution density of the energy parameter;

[0035] The amplitude threshold interval and the phase correlation value are logically combined to generate a composite rule including multiple logical operators and time window constraints.

[0036] Optionally, the performing multi-band processing on the voltage fluctuation signal to generate voltage segment features and recording the amplitude parameters of the voltage fluctuation signal; performing feature extraction on the transient signal to generate a transient peak by capturing energy density changes in a region where the signal amplitude changes suddenly, and recording the energy parameters of the transient signal include:

[0037] The voltage fluctuation signal is divided into multiple basic units according to a preset time length. The frequency range of each basic unit is divided based on the fundamental frequency and harmonic frequency distribution range of the power line. The energy distribution value of the signal in each frequency range is counted and accumulated to form the voltage segment feature.

[0038] detecting a maximum value and a minimum value of a signal amplitude in each basic unit according to the voltage segment characteristics, and taking a difference between the maximum value and the minimum value as an amplitude parameter corresponding to the basic unit;

[0039] A sliding window is used to intercept the area in the transient signal where the amplitude change rate exceeds a preset condition as the amplitude mutation area, and the point of amplitude mutation in the amplitude mutation area is identified as the critical point. The energy density value corresponding to the critical point is taken as the transient peak value, and the total energy of the amplitude mutation area is accumulated to generate the energy parameter.

[0040] In a second aspect, the present application provides an intelligent fault diagnosis system for an intelligent pod, comprising:

[0041] An acquisition module is used to obtain voltage fluctuation signals of the smart pod power line and transient signals of the motor control line;

[0042] a processing module, configured to perform multi-band processing on the voltage fluctuation signal, generate voltage segment features, and record amplitude parameters of the voltage fluctuation signal; perform feature extraction on the transient signal, generate transient peaks by capturing energy density changes in regions where the signal amplitude changes suddenly, and record energy parameters of the transient signal;

[0043] An analysis module is configured to establish a signal trigger point synchronization mechanism based on the voltage segment characteristics and the transient peak value, and to perform a comparative analysis of the voltage fluctuation amplitude and the transient amplitude based on the synchronization trigger point, and calculate the time difference between the two;

[0044] A generating module, configured to generate a composite rule including an amplitude threshold interval and a phase correlation value based on a preset periodic reference signal in combination with the amplitude parameter and the energy parameter;

[0045] An establishment module is used to construct a multi-dimensional space including the amplitude parameter, the energy parameter and the time difference, and establish a fault type prediction model in the multi-dimensional space based on the composite rule;

[0046] The prediction module is used to identify abnormal points by calculating the matching degree between the distribution position of the feature vector in the multi-dimensional space and the fault type prediction model, and determine the corresponding fault type according to the distribution pattern of the abnormal points in the multi-dimensional space, and output a diagnostic result including the identified fault type and the corresponding abnormal point.

[0047] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an intelligent fault diagnosis method for an intelligent pod as described in the first aspect above.

[0048] In a fourth aspect, the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements an intelligent fault diagnosis method for an intelligent pod as described in the first aspect.

[0049] In the example of the present application, the voltage fluctuation signal of the power line of the intelligent pod and the transient signal of the motor control line are obtained; the voltage fluctuation signal is subjected to multi-band processing to generate voltage segment characteristics, and the amplitude parameters of the voltage fluctuation signal are recorded; the transient signal is subjected to feature extraction, and the transient peak value is generated by capturing the energy density change in the signal amplitude mutation area, and the energy parameters of the transient signal are recorded; based on the voltage segment characteristics and the transient peak value, a signal trigger point synchronization mechanism is established, and based on the synchronization trigger point, a comparative analysis of the voltage fluctuation amplitude and the transient amplitude is performed to calculate the time difference between the two; based on the preset A periodic reference signal is combined with the amplitude parameter and the energy parameter to generate a composite rule including an amplitude threshold interval and a phase correlation value; a multidimensional space including the amplitude parameter, the energy parameter and the time difference is constructed, and a fault type prediction model is established in the multidimensional space based on the composite rule; abnormal points are identified by calculating the matching degree between the distribution position of the feature vector in the multidimensional space and the fault type prediction model, and the corresponding fault type is determined according to the distribution pattern of the abnormal point in the multidimensional space, and a diagnostic result including the identified fault type and the corresponding abnormal point is output.

[0050] The technical solution of this application has the following beneficial effects:

[0051] This application synchronously collects the voltage fluctuation signal of the smart pod power line and the transient signal of the motor control line, performs multi-band decomposition on the voltage fluctuation signal to generate voltage segment characteristics and records the amplitude parameters, and simultaneously performs energy density analysis on the mutation area of ​​the transient signal to generate transient peak and energy parameters, establishes a signal trigger point synchronization mechanism to realize the time series comparison analysis of the voltage fluctuation amplitude and the transient amplitude and calculates the time difference, and generates a composite rule containing amplitude threshold interval and phase correlation value based on the periodic reference signal fusion amplitude parameter and energy parameter, constructs a spatial model integrating multi-dimensional parameters and establishes a fault prediction rule library, and finally identifies the distribution pattern of abnormal points through the distribution matching degree of feature vectors in space, realizes millisecond-level precise positioning of typical faults such as power system harmonic interference, motor drive transient overload, and control signal distortion, solves the technical problems such as insufficient cross-domain correlation analysis of multi-source heterogeneous signals, poor adaptability to dynamic environment and high misjudgment rate of composite faults, and significantly improves the fault diagnosis and accuracy of smart pods under complex working conditions.

[0052] Furthermore, the feature vector is converted to a multi-dimensional space through a nonlinear coordinate mapping function to generate coordinate points. After determining the abnormal points based on the dynamic threshold of geometric distance, a dual filtering mechanism of spatial aggregation and temporal continuity is implemented to screen the effective abnormal point set, and then its distribution geometric characteristics and time series characteristics are extracted to map the fault type label. The final package is output as a standardized diagnostic data packet that complies with the Industrial Internet of Things protocol. This method breaks through the limitations of traditional linear diagnosis through nonlinear mapping, uses a dynamic threshold mechanism to adapt to complex working conditions, and a dual filtering mechanism to effectively suppress environmental noise interference. The characteristic distribution pattern analysis realizes accurate mapping of fault types, solving technical problems such as insufficient multi-dimensional signal correlation analysis, poor dynamic environment adaptability, and high compound fault misjudgment rate. It significantly improves the diagnostic reliability of the smart pod in severe vibration and electromagnetic interference environments. At the same time, it achieves seamless connection with the Industrial Internet of Things platform through standardized data packaging, providing a reliable decision-making basis for equipment health management.

[0053] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0055] Figure 1 A flow chart of an intelligent fault diagnosis method for an intelligent pod provided by the present application is shown;

[0056] Figure 2 A scenario diagram of an intelligent fault diagnosis method for an intelligent pod provided by the present application is shown;

[0057] Figure 3 A schematic diagram of the structure of an intelligent fault diagnosis system for an intelligent pod provided by the present application is shown;

[0058] Figure 4 A schematic structural diagram of a computing device provided by the present application is shown. DETAILED DESCRIPTION

[0059] In order to enable people skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0060] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0061] Research shows that drone inspections and smart pods in ship power systems need to achieve millisecond-level coordinated control of multiple subsystems such as power supplies, motors, and communications under complex electromagnetic environments and severe vibration conditions. Their fault diagnosis technology requires the construction of a full-link monitoring system and dynamic diagnosis algorithms. However, existing solutions based on fault tree analysis have defects such as insufficient identification of complex faults, weak cross-domain correlation of multi-source data, and poor dynamic adaptability of offline models. These defects result in a low detection rate for new cascading failures and a significant decrease in diagnostic accuracy with environmental fluctuations, which seriously restricts the reliability of continuous operations under complex working conditions.

[0062] To address these issues, the present invention proposes an intelligent fault diagnosis method for smart pods. Its core approach involves synchronously acquiring power line voltage fluctuation signals and motor control line transient signals. This method generates voltage segment features through multi-band decomposition and extracts transient peak energy density parameters. A signal trigger point synchronization mechanism is established to enable time-series comparative analysis of voltage fluctuation amplitudes and transient amplitudes. Combining this with a periodic reference signal, a composite rule is constructed that incorporates amplitude threshold intervals and phase correlation values. Furthermore, a multidimensional spatial model is constructed that integrates amplitude parameters, energy parameters, and time differences. Within this spatial model, the distribution pattern of anomalies is identified through the matching degree of feature vector distributions. Ultimately, a diagnostic result is output that includes the location of the fault type and the spatiotemporal coordinates of the anomaly points. This method breaks through the limitations of traditional linear diagnosis through nonlinear coordinate mapping, uses a dynamic threshold mechanism to adapt to complex working conditions, uses a time-space dual filtering mechanism to effectively suppress environmental noise interference, and uses characteristic distribution pattern analysis to achieve accurate identification of new composite faults such as the superposition of power supply harmonic interference and control signal jitter. It solves the core problems of the existing technology, such as the lack of cross-domain correlation analysis, the lag in dynamic threshold adjustment, and the lack of collaborative modeling of multi-source data. It improves the accuracy of composite fault identification in the scenarios of drone steering control and ship pod drive, shortens the diagnostic response time, and improves the autonomous diagnosis reliability and task execution continuity of the intelligent pod in complex electromagnetic environments.

[0063] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0064] Figure 1 The present invention provides a flowchart of an intelligent fault diagnosis method for an intelligent pod, as shown in FIG. Figure 1 As shown, the method includes:

[0065] 101. Obtain a voltage fluctuation signal of the smart pod power line and a transient signal of the motor control line;

[0066] In the above scheme, the voltage fluctuation signal refers to the periodic or non-periodic change signal of the voltage amplitude in the power line of the intelligent pod caused by load changes, electromagnetic interference or power module abnormality. It contains the power frequency fundamental component and high-frequency harmonic components, reflecting the transient response characteristics of the power supply system; the transient signal refers to the nanosecond spike pulse signal generated by the action of the power switching device during the start-stop and speed regulation process of the motor control line. It contains characteristic parameters such as energy density and rising edge steepness, and characterizes the transient stress state of the drive circuit.

[0067] In an embodiment of the present application, through step 101, a high-precision differential probe and an isolation amplifier are respectively deployed at the output end of the power line and the input end of the motor control line, and the timing alignment of the dual-channel signals is ensured through a synchronous clock source to collect the original analog voltage fluctuation signal and the transient impact signal.

[0068] In practical applications, a 200MHz differential probe with a 1MΩ differential input impedance and a 120dB common-mode rejection ratio is connected in series to the power line output. This probe captures voltage fluctuations in the 0.1Hz-10MHz range. An isolation amplifier is connected in parallel to the motor control line input, utilizing magnetic isolation technology to achieve 2500Vrms electrical isolation between the input and output. This amplifier has a bandwidth of 50MHz and is used to capture transient impulse signals in the 10kHz-50MHz range during motor startup and shutdown. A high-precision clock source generates a synchronized 100MHz clock signal for both channels, ensuring a phase deviation of less than 1ns between the dual-channel sampling clocks.

[0069] The above-mentioned 101 overall solution, through the wide-band signal synchronous acquisition technology, completely preserves the physical characteristics of the steady-state fluctuations and transient shocks of the power supply system, realizes high-precision spatiotemporal correlation analysis of multi-dimensional signals, significantly improves the ability to trace the abnormalities of the pod energy link under complex working conditions, and provides a complete data foundation for system reliability optimization.

[0070] 102. Perform multi-band processing on the voltage fluctuation signal to generate voltage segment features, and record the amplitude parameters of the voltage fluctuation signal; perform feature extraction on the transient signal to generate transient peaks by capturing energy density changes in signal amplitude mutation regions, and record the energy parameters of the transient signal;

[0071] Optionally, step 102 may specifically include the following steps:

[0072] 1021. Divide the voltage fluctuation signal into a plurality of basic units according to a preset duration, divide each basic unit into a frequency range based on the fundamental frequency and harmonic frequency distribution range of the power line, calculate the energy distribution value of the signal within each frequency range, and accumulate the energy distribution values ​​within each frequency range to form a voltage segment feature;

[0073] 1022. Detect a maximum value and a minimum value of a signal amplitude in each basic unit according to the voltage segment characteristics, and use a difference between the maximum value and the minimum value as an amplitude parameter corresponding to the basic unit;

[0074] 1023. Use a sliding window to intercept the area in the transient signal where the amplitude change rate exceeds a preset condition as the amplitude mutation area, identify the point of amplitude mutation in the amplitude mutation area as the critical point, use the energy density value corresponding to the critical point as the transient peak value, and accumulate the total energy of the amplitude mutation area to generate an energy parameter.

[0075] In the above scheme, the voltage segment feature refers to the power signal spectrum distribution vector generated by the accumulation of multi-band energy, which can be used to evaluate the degree of harmonic pollution in the power supply system. The amplitude parameter refers to a quantitative indicator reflecting the intensity of voltage fluctuations, which can be used to identify transient load anomalies in the power supply line. The amplitude mutation area refers to the local interval where the rate of change in the transient signal exceeds the threshold, which can be used to locate the switching transient events of the drive circuit. The critical point refers to the position of the extreme value of the signal change rate in the amplitude mutation area, which can be used to calibrate the core occurrence time of the transient impact. The transient peak refers to the transient energy density value corresponding to the critical point, which can be used to quantify the transient stress level of the power device. The energy parameter refers to the total energy integral value in the amplitude mutation area, which can be used to evaluate the impact resistance of the drive circuit.

[0076] In the embodiment of the present application, the voltage fluctuation signal processing is started at step 1021, and the original signal is divided into continuous basic units according to the preset time length; after each unit is weighted by the Hanning window, it is converted into a spectrum through the fast Fourier transform; based on the power supply fundamental frequency, the power supply fundamental frequency is obtained. , using an adaptive frequency band division strategy to monitor electromagnetic interference, the frequency band division strategy covers the low and medium frequency bands Fundamental wave and low harmonics, mid-frequency band Capture switching noise, high frequency band ; Calculate the energy value of each frequency band and accumulate it to generate a three-dimensional voltage segment feature vector , which is used to characterize the energy aggregation characteristics of different frequency bands.

[0077] Then, the frequency band analysis result in step 1021 is continued through step 1022. When the medium-, medium-, and high-frequency energy increases abnormally, the time domain waveform of the corresponding basic unit is automatically locked. A dual-pointer sliding window extreme value detection algorithm is used to scan the time domain signal with a window length of 1 / 4 fundamental wave period, and the maximum and minimum values ​​within the window are dynamically recorded. The amplitude parameter is calculated as a quantitative indicator of the voltage fluctuation intensity to reflect the transient load impact intensity of the power supply line. The calculation formula is as follows: ,in is the amplitude parameter, is the maximum value in the dynamic recording window, is the minimum value in the dynamic recording window.

[0078] Finally, the transient signal is processed in parallel through step 1023, and the first-order derivative of the current signal is calculated through the filter. The derivative signal is scanned using a sliding window of width. When the first-order derivative in the window exceeds the preset threshold, the window is marked as an amplitude mutation area; the second-order derivative zero point is located in the amplitude mutation area as the critical point, and according to the formula Extract the transient peak value of this point, where is the current corresponding to the critical point, is the sampling interval; the energy parameter is generated by integrating the signal in the mutation area, and the calculation formula is as follows: ,in and are the upper and lower bounds of the mutation region.

[0079] In actual applications, in a sudden acceleration scenario of the gimbal of a six-axis drone intelligent pod, the voltage fluctuation signal of the 48V power supply line is divided into basic units according to the duration of 10ms. After Hanning window weighting and fast Fourier transform, the frequency band is divided based on the 50Hz base frequency. The low frequency range is 0-150Hz, with an energy share of 65%. The mid-frequency range is 150Hz-1kHz, and an abnormal energy surge is detected. The high frequency range is 1kHz-500kHz, showing a 750kHz switching ripple. Execution is performed on the abnormal unit. A dual-pointer sliding window extreme value detection captures the maximum value of 50.2V and the minimum value of 48.7V in the time domain waveform, generating an amplitude parameter of 1.5V. Simultaneously, the synchronous motor control line transient signal is processed. After calculating the current derivative using a Savitzky-Golay filter, a 1μs sliding window scan is used to identify regions of amplitude mutation. The critical point is located at t=23.6ms, and the transient peak value of 25kW / μs is extracted. The current signal in this region with a width of 0.9μs is integrated to generate an energy parameter of 1.8mJ.

[0080] The 102-point overall solution, described above, generates spectral distribution features by applying multi-band energy accumulation to voltage fluctuation signals, simultaneously extracting time-domain amplitude fluctuation parameters to achieve quantitative stratification of power supply harmonic characteristics. Furthermore, transient signal differential analysis accurately captures regions of amplitude mutation, combining critical point energy density with regional total energy integration to deconstruct the core characteristics driving transient events. This solution transcends the limitations of traditional single-dimensional diagnostics by constructing a composite feature system that combines frequency domain breadth with time domain precision. This provides multi-scale criteria for tracing anomalies and assessing the health status of pod energy links, significantly enhancing system reliability analysis capabilities under complex operating conditions.

[0081] 103. Based on the voltage segment characteristics and the transient peak value, establish a signal trigger point synchronization mechanism, and based on the synchronization trigger point, perform a comparative analysis of the voltage fluctuation amplitude and the transient amplitude, and calculate the time difference between the two;

[0082] Optionally, step 103 may specifically include the following steps:

[0083] 1031. Generate a dual constraint condition based on the amplitude duration and the amplitude change slope, screen the voltage abnormal signal segments whose amplitude parameters exceed a first threshold in the voltage segment characteristics based on the dual constraint condition, and locate the energy mutation region where the energy parameter mutation exceeds a second threshold in the transient peak based on the consistency of timestamp alignment and energy density gradient direction;

[0084] 1032. Establish a signal trigger point synchronization mechanism to compare the time overlap between the abnormal voltage signal segment and the energy mutation region point by point, and select matching events with timestamp differences less than a preset tolerance and mark them as synchronization trigger points.

[0085] 1033. Taking the synchronous trigger point as a reference, calculate the amplitude change slope of the voltage fluctuation signal before and after the trigger, and simultaneously calculate the energy density change rate of the transient peak before and after the trigger, and determine the time difference between the voltage fluctuation amplitude and the starting moment of the transient amplitude based on the intersection of the amplitude change slope and the energy density change rate.

[0086] In the above scheme, dual constraints refer to logical rules for screening power anomaly events. They include parallel criteria for a lower limit on the duration of the amplitude and an upper limit on the slope of change. These conditions can be used to accurately locate complex fault characteristics characterized by both persistent and sudden changes in power systems. A voltage anomaly signal segment refers to a voltage fluctuation on a power line that significantly deviates from steady-state conditions. It includes time-domain waveform segments with amplitude exceeding a threshold, duration exceeding a threshold, and a limited rate of change. These can be used to diagnose abnormal load response or electromagnetic compatibility failures in power modules. An energy abrupt change region refers to a transient pulse region in a motor control signal where energy density changes dramatically. These regions include local signal features such as energy gradient reversal and amplitude exceeding a threshold. These can be used to identify power device switch overload or driver logic errors. A synchronous trigger point is a reference time anchor for cross-signal domain correlation events. It includes matching time labels for the voltage anomaly segment and the energy abrupt change region, which are spatiotemporally aligned. This allows for establishing a causal temporal chain between power disturbances and motor responses. A time difference refers to the time delay between the start of the voltage fluctuation and the start of the transient response. This time difference includes a conduction delay quantified by the slope intersection method and can be used to locate signal transmission bottlenecks or fault propagation paths in energy links.

[0087] In the embodiment of the present application, first, a dual constraint screening is applied to the voltage segment characteristics in step 1031, and the energy accumulation value of the voltage signal is calculated by the sliding window integration algorithm to screen out the continuous abnormal segments whose energy exceeds the first threshold; at the same time, the central difference method is used to calculate the instantaneous slope, and the slowly varying segments whose rate of change is lower than the preset upper limit are retained; for the transient peak signal, the Sobel operator is used to calculate the energy density gradient , locate the energy mutation area where the gradient direction changes suddenly and the amplitude exceeds the second threshold; the two types of signals are roughly aligned in the time domain through the dynamic time warping algorithm. Taking the voltage abnormal section as the benchmark, the transient signal time axis is dynamically stretched or compressed to eliminate the millisecond-level deviation caused by the system clock drift, and the candidate event pairs with timestamp alignment are output.

[0088] Then, in step 1032, based on the alignment result in step 1031, high-precision matching is performed in the time overlap interval between the voltage abnormal segment and the energy mutation region; the dual-channel data is resampled with 100ps resolution to construct a matching function based on cosine similarity. ,in represents the normalized vector of the voltage abnormal signal segment at time point t, Represents the normalized vector of the energy mutation area at time point t; when 5 consecutive sampling points are detected to meet When the time difference is greater than a certain value and the time deviation is less than a certain value, it is determined to be a valid event association. Take the time median of the continuous interval It serves as a synchronization trigger point and is accompanied by a confidence score, for example, 0.98 indicates 98% matching reliability.

[0089] Finally, step 1033 is performed to synchronize the trigger point Start transient response analysis for the time origin. Within a ±100μs window, a linear regression fit is performed on the voltage fluctuation signal to obtain the amplitude change slope. A second-order polynomial fit is performed on the transient energy density and its derivative is taken to generate the instantaneous rate of change function. The intersection moment is solved by analytically solving the simultaneous equations and the conduction time difference is calculated.

[0090] In actual application, a voltage drop of -1.5V and 18ms duration was detected in the 48V DC bus of the power supply line. At the same time, the Sobel operator was used to locate the energy mutation area where the energy density gradient of the motor control line suddenly changed to +0.8J / μs at 12.584s. The two types of events were roughly aligned to the time window of [12.580s, 12.598s] using the dynamic time warping algorithm. Then, the dual-channel data was resampled with a resolution of 100ps, and the cosine similarity was calculated in the overlapping window. Seven consecutive points that met the requirements were captured at 12.5912s. And the time deviation is less than 0.3ns, determine the synchronization trigger point ; Finally As a benchmark, the linear fitting voltage fluctuation is used to obtain a slope of -28V / ms. The second-order polynomial is used to fit the transient energy function and the derivative is used to generate the instantaneous rate of change function. The instantaneous rate of change is recorded as , the intersection point is obtained by solving the simultaneous equations , the final time difference is calculated as It shows that the transient response lags behind the power supply fluctuation, which is located as a conduction delay feature caused by the discharge delay of the energy storage capacitor.

[0091] The 103-point overall solution, described above, uses dual constraints to precisely screen power anomalies and energy mutation events, combining dynamic time warping to achieve millisecond-level time-domain coarse alignment. High-resolution resampling and cosine similarity matching are used to establish nanosecond-level synchronization trigger points, breaking through the timing accuracy bottleneck of traditional signal correlation. A slope-rate-of-change coupled analytical method is used to quantify the conduction delay of voltage fluctuations and transient responses, revealing the dynamic process of abnormal conduction in the energy chain. This entire process forms a closed-loop diagnostic chain, from event detection, spatiotemporal synchronization, to delay analysis. This significantly improves the pod system's fault tracing capabilities under complex operating conditions, provides a verifiable temporal evidence foundation for multi-physics field coupling analysis, and comprehensively enhances the controllability and reliability of the energy chain.

[0092] 104. Based on a preset periodic reference signal, in combination with the amplitude parameter and the energy parameter, generate a composite rule including an amplitude threshold interval and a phase correlation value;

[0093] Optionally, step 104 may specifically include the following steps:

[0094] 1041. Synchronously divide the preset periodic reference signal into a plurality of periodic units of equal length according to the power frequency period of the power supply line;

[0095] 1042. Within each of the periodic units, set the allowable fluctuation range of the amplitude parameter based on the statistical distribution of historical normal operating condition data to generate an amplitude threshold interval;

[0096] 1043. Determine, based on the distribution density of the energy parameter, a phase correlation value between the energy parameter and the start time of the periodic unit;

[0097] 1044. Logically combine the amplitude threshold interval and the phase correlation value to generate a composite rule including multiple logical operators and time window constraints.

[0098] In the above scheme, the periodic reference signal refers to a synchronous timing scale generated with the power frequency as a reference, which includes the characteristics of equally spaced periodic division units and has the characteristics of a timing framework with strict phase alignment, and can be used to implement phase domain correlation analysis of multi-source signals. The amplitude threshold interval refers to the allowable range of voltage fluctuations based on historical operating condition statistics, and has the time-varying characteristics of adaptive adjustment with the power frequency phase, and can be used to identify abnormal power amplitude fluctuation events. The phase correlation value refers to the relative position of the transient energy feature within the periodic unit, and has the probability density distribution focusing characteristics, which can be used to determine the coupling relationship between transient events and power phases. The composite rule refers to the multi-dimensional judgment logic that integrates amplitude and phase characteristics, includes the characteristics of time-space correlation constraints, has the diagnostic characteristics of enhanced anti-interference, includes logical operators and time window constraint parameters, and can be used to achieve highly robust energy link anomaly judgment.

[0099] In the embodiment of the present application, first, in step 1041, a digital phase-locked loop chip is used to capture the power supply frequency signal and generate a phase-locked periodic reference signal. A programmable frequency divider is used to divide each power frequency cycle into 20 1ms period units, and the frequency division coefficient strictly follows the formula Dynamic calibration. Adaptive clock compensation maintains cell boundary jitter below 10ns when power frequency fluctuates within ±0.5Hz. The output cell index sequence carries a globally synchronized timestamp, establishing a precise phase coordinate system for subsequent analysis.

[0100] Next, in step 1042, statistical modeling is performed within each cycle unit based on the normal operating condition samples in the historical database. The amplitude parameter sequence corresponding to the current unit index is extracted, and the sliding mean μ and standard deviation σ are calculated using the Welford recursive algorithm. A basic threshold interval is then generated based on the μ±3σ principle. A cubic spline function is used to smooth the threshold transitions between adjacent units. For special operating conditions, such as the motor acceleration phase, the threshold bandwidth is compressed by 5%. The generated dynamic threshold curve is stored in a rule base, and its phase alignment ensures a strict match with the signal.

[0101] Then, in step 1043, the probability density modeling of the transient energy parameters is performed in each period unit, and the density estimator is constructed using the Epanechnikov kernel function. , where the bandwidth h is optimized to 0.05ms according to the Silverman criterion; the peak moment of positioning density Then, calculate its relative to the starting point of the unit The normalized phase offset is calculated as follows: Based on cluster analysis of historical normal data, a valid phase interval is generated, such as the motor commutation phase [100°, 120°]. This phase correlation value quantifies the characteristic position of the transient event in the power frequency cycle.

[0102] Finally, multi-dimensional feature fusion is implemented in the configurable rule engine through step 1044, and the dynamic threshold curve and phase interval are bound to the corresponding unit index. Then, a tree logic structure is constructed. Crucially, a time window constraint is introduced to eliminate sporadic interference, for example, requiring the abnormal state to persist for more than three units. Finally, an optimized decision circuit is generated through the hardware description language and burned into the FPGA to achieve 5μs-level diagnosis.

[0103] In a practical application, during a full-load acceleration test of the gimbal motor on a six-axis mapping drone pod, a 50Hz power frequency signal from the power management unit is input to the PC1 phase comparator of a CD4046BE phase-locked loop (PLL). A second-order active loop filter suppresses frequency jitter, and the resulting synchronization reference signal is fed into the global clock network of a Xilinx Artix-7 FPGA. The FPGA is internally configured with a digital frequency division chain, operating at a main frequency of 100MHz ±5ppm and a division factor of 100,000. This generates 20 strictly phase-aligned 1ms periodic units. The third unit, corresponding to the time window [1.000ms, 2.000ms], is labeled "motor start phase." This label is transmitted to a DDR3 historical database via the AXI bus. The historical database retrieves the peak voltage data of the most recent 10,240 normal operating conditions for Unit 3 and uses the fixed-point Welford recursive algorithm to update the statistical parameters: the mean μ is 47.025V, the standard deviation σ is 0.479V, and the dynamic threshold interval [45.588V, 48.462V] is generated. To avoid threshold jumps between adjacent units, Catmull-Rom spline interpolation is used with a tension coefficient of 0.5 to achieve a smooth transition. The transient energy signal within the Unit 3 time window is synchronously collected and input into the AD8479 isolation amplifier through an anti-aliasing filter to obtain a sequence of pulse integral values. Kernel density estimation is performed on this sequence, and the Epanechnikov kernel function generates a probability density curve. , locate the main peak time , the relative unit starting point is offset by 0.283ms, and the phase correlation value is calculated Combined with the historical data DBSCAN clustering, the effective phase interval [100.2°, 119.7°] was determined.

[0104] The above-mentioned 104 overall solution achieves a systematic upgrade of the intelligent pod energy link anomaly detection capability by constructing a composite diagnostic rule that couples dynamic thresholds with phase characteristics. Based on the spatiotemporal reference framework of the power frequency cycle, the solution first establishes a dynamic threshold range for the voltage amplitude parameter. Its boundary is adaptively adjusted according to the statistical characteristics of historical operating conditions, effectively overcoming the adaptability defects of traditional fixed thresholds to complex operating conditions. The synchronously extracted transient energy phase correlation value accurately binds the signal characteristics to the power frequency phase coordinate system, breaking through the limitations of single amplitude detection for the lack of timing information. The multi-dimensional composite rule formed by integrating the amplitude threshold, phase interval and time window constraints through logical operators significantly enhances the anti-interference robustness of the diagnostic system and can accurately identify the abnormal coupling pattern of power supply fluctuations and motor transients in the three-dimensional space of amplitude, phase and time.

[0105] 105. Construct a multidimensional space including the amplitude parameter, the energy parameter, and the time difference, and establish a fault type prediction model in the multidimensional space based on the composite rule;

[0106] Optionally, step 105 may specifically include the following steps:

[0107] 1051. Arrange the amplitude parameter, the energy parameter, and the time difference in chronological order into a three-dimensional data unit including a timestamp identifier, and construct a multi-dimensional space;

[0108] 1052. Based on the composite rule and in combination with geometric distribution characteristics of the three-dimensional data units in the multi-dimensional space, divide the multi-dimensional space into cluster regions that satisfy the composite rule;

[0109] 1053. Perform statistical analysis on the cluster area, extract the cluster center coordinates, cluster range boundary values, and density distribution characteristics of the coordinate points within the cluster area to form cluster feature data;

[0110] 1054. Match the cluster feature data with typical fault features in a pre-stored fault type library to generate a fault type prediction model including cluster center coordinate weights, dynamic adjustment coefficients of cluster range boundaries, and density distribution matching thresholds.

[0111] In the above scheme, multidimensional space refers to a feature mapping space composed of amplitude parameters, energy parameters, and time differences, including nonlinear correlation topological structures between parameters, which is used to quantify the distribution patterns of fault characteristics. The fault type prediction model refers to a feature matching engine dynamically generated based on composite rules, including three core parameters: cluster center weight, boundary adjustment coefficient, and density matching threshold, to achieve probabilistic identification of fault modes. The cluster region refers to a data aggregation domain in the multidimensional space that meets the constraints of composite rules. Its geometric shape reflects the consistency of the physical representation of a specific fault and is used to define the effective range of the fault judgment criteria. Cluster feature data refers to a tuple that describes the statistical characteristics of the cluster region, including the coordinates of the cluster center that characterize typical fault states, the boundary value matrix that defines the fault judgment criteria, and the density distribution function that reflects the probability of fault occurrence. The fault type library refers to a pre-existing knowledge base of typical fault characteristics, including cluster feature data and diagnostic rules of historical fault cases, providing a benchmark comparison framework for fault prediction.

[0112] In the embodiment of the present application, the amplitude parameters of the voltage fluctuation signal, the energy parameters of the transient signal, and the time difference between events are first strictly aligned according to nanosecond timestamps in step 1501. Then, the sampling gaps are filled using a cubic spline interpolation algorithm to generate a dense three-dimensional data stream. This data stream is mapped to a multidimensional space consisting of an amplitude axis, an energy axis, and a time difference axis. Its topological structure retains key nonlinear correlations through manifold dimensionality reduction technology. For example, when the time difference approaches zero, the coupling relationship between the amplitude drop and the energy pulse appears as a twisted spiral trajectory in space, laying a geometric foundation for subsequent fault analysis.

[0113] Next, in step 1502, a constrained hypersurface is dynamically constructed based on pre-set composite rules. A KD-Tree spatial index is used to quickly identify clusters of data points that meet the initial screening criteria. The OPTICS density clustering algorithm is then used to further explore the spatial structure. By calculating core distance and reachable distance, densely connected regions are identified. Each connected region is labeled as a cluster, and its spatial form directly reflects the physical nature of the fault. For example, an ellipsoidal cluster with amplitude parameters of [36V, 42V], energy parameters of [0.25J, 0.35J], and inter-event time differences of [0.2ms, 0.4ms] is detected, which is mapped to the fault characteristic domain of "IGBT overcurrent breakdown."

[0114] Then, a triple quantitative analysis is performed on each cluster area. The cluster center coordinates are calculated using the Mahalanobis distance weighted mean method. ; Extract cluster range boundaries through α-shape non-convex envelope algorithm , accurately capture the concave and convex features of the boundary; use adaptive bandwidth kernel density estimation to construct the density distribution function ;Finally output cluster feature data tuple ,Spatial statistical characteristics of complete package failures.

[0115] Finally, in step 1504, the cluster feature data is multimodally matched with the fault type library, and dynamic time warping is used to align the current cluster center with the trajectory of the historical fault, and the similarity weight is calculated. ; Adjust the pre-stored fault boundary matrix B through the gradient descent optimizer to generate dynamic coefficients To adapt to new working conditions; evaluate the current density distribution based on KL divergence Deviation from typical fault, output matching threshold Finally, a fault type prediction model is generated, whose output is a triplet , realizing the probabilistic classification of faults.

[0116] In practical applications, in the diagnosis of the motor drive system of a six-axis agricultural drone pod, a high-precision synchronous acquisition module was used to obtain the voltage fluctuation signal and transient current of the motor control line during the servo operation, and extract the amplitude parameters, energy parameters, and event time difference. After aligning the 100MHz timestamp, a three-dimensional data unit sequence (41.5V, 0.38J, 0.28ms) was generated through cubic spline interpolation and mapped to the amplitude-energy-time difference space. Its manifold structure shows that the amplitude drop and energy pulse are in the same direction. The region forms a high curvature spiral cluster. Clustering is initiated based on the composite rule and the KD-Tree index is used to 327 candidate points were selected from the data points, and the OPTICS algorithm identified the core density area. The clustering domain is distributed in an ellipsoidal shape with a magnitude of ,energy , time difference The geometric features are consistent with the IGBT overcurrent breakdown mode. The domain is quantified and the cluster center is calculated using Mahalanobis distance weighting. , α-shape algorithm extracts non-convex boundaries , kernel density estimation constructs a bimodal density function .Will Matching with the fault database, DTW is used to align the current center trajectory with the historical "IGBT breakdown" trajectory. Gradient descent optimizes the boundary matrix to obtain a dynamic coefficient α = 1.08, and the KL divergence calculation density matching threshold β = 0.89. Finally, a fault prediction model is generated. , diagnose new data points Output "IGBT breakdown" probability.

[0117] The overall solution of 105 mentioned above realizes high-dimensional geometric representation of fault signals by constructing a multi-dimensional feature space, laying a structural foundation for abnormal pattern recognition under complex working conditions. Based on the adaptive clustering mechanism driven by composite rules, clustering areas reflecting the physical nature of specific faults are accurately divided, significantly improving the pertinence and completeness of fault feature extraction. Relying on multi-scale statistical analysis of cluster feature data, a holographic fault fingerprint library is formed that integrates typical state coordinates, dynamic boundary constraints and probability density distribution. Finally, through intelligent matching of the fault type library and dynamic calibration of parameters, a prediction model with strong generalization ability is generated, which not only realizes probabilistic and accurate diagnosis of fault types, but also breaks through the rigid limitations of traditional threshold criteria, providing predictive maintenance decision support for smart pods that is comprehensive, robust and explainable.

[0118] 106. Identify abnormal points by calculating the matching degree between the distribution position of the feature vector in the multi-dimensional space and the fault type prediction model, and determine the corresponding fault type according to the distribution pattern of the abnormal points in the multi-dimensional space, and output a diagnostic result including the identified fault type and the corresponding abnormal point.

[0119] Optionally, step 106 may specifically include the following steps:

[0120] 1061. Convert the feature vector into a coordinate point in the multidimensional space using a preset nonlinear coordinate mapping function, calculate the geometric distance between the coordinate point and the historical benchmark data form in the fault type prediction model, and mark the corresponding coordinate point whose geometric distance exceeds a dynamic threshold as an abnormal point;

[0121] 1062. Apply a dual filtering mechanism including spatial filtering and temporal filtering to the outliers to screen out outliers that meet both spatial aggregation and temporal continuity characteristics to generate an outlier set;

[0122] 1063. Input the abnormal point set into the regional feature analysis module, extract the distribution geometric features and time series features of the abnormal point set, map the fault type labels according to the distribution geometric features and time series features, and assign them to the corresponding abnormal points;

[0123] Among them, step 1063 may specifically include the following processes: in the regional feature analysis module, the coordinate span of the abnormal point set in the three-dimensional space is counted to determine the spatial distribution range, and the average distance between adjacent points is calculated to evaluate the spatial discreteness of the abnormal point set, and the spatial distribution range and the spatial discreteness are used as distribution geometric features; at the same time, the frequency distribution of abnormal points generated in the abnormal point set within unit time is counted, and a time interval sequence is generated according to the difference between the timestamps of adjacent abnormal points, and the frequency distribution and the time interval sequence are combined as time series features; the spatial distribution feature and the time series feature are fused to form a composite feature descriptor, and the composite feature descriptor is matched with the historical fault samples in the fault type prediction model for similarity, and the fault type label corresponding to the matched fault mode is assigned to the corresponding abnormal point through feature similarity calculation.

[0124] 1064. Encapsulate the fault type label and the timestamp and spatial coordinates of the corresponding abnormal point into a diagnostic data packet that complies with the Industrial Internet of Things protocol to output a diagnostic result.

[0125] In the above scheme, anomaly points refer to data coordinate points that deviate from the historical benchmark distribution in the multi-dimensional feature space, which can be used to identify potential abnormal states of the system. The dynamic threshold refers to the distance judgment threshold that is adaptively adjusted according to the system operating conditions, including the baseline fluctuation range and the environmental interference tolerance, which can be used to filter out false triggers caused by random noise. The dual filtering mechanism refers to the joint screening logic that combines spatial aggregation and temporal continuity, including local density criteria and continuous occurrence frequency conditions, which can be used to improve the credibility of abnormal events. The anomaly point set refers to the group of spatiotemporal correlated anomaly points generated by the filtering mechanism. The composite feature descriptor refers to the feature vector that integrates the geometric distribution characteristics and the temporal evolution law, including the spatial discreteness index and the event interval statistics, which can be used to map the essential characteristics of the fault mode. The diagnostic result refers to the structured output data containing the fault type label and traceability basis, including spatial positioning information and the initial trigger timestamp, which can be used to guide precise maintenance decisions.

[0126] In the embodiment of the present application, the feature vector is first projected into a three-dimensional multi-dimensional space through a nonlinear coordinate mapping function to generate coordinate points with physical meaning. The Mahalanobis distance between this point and the historical fault cluster center stored in the fault type prediction model is calculated. The specific calculation formula is as follows: ,in is the centroid of the fault. The system dynamically calculates the distance threshold is the mean distance of the normal working condition samples, where is the standard deviation. When the distance from P to the nearest cluster center When , mark it as the initial outlier point and record its spatial coordinates and timestamp This step transforms the abstract features into quantifiable spatial deviation indicators.

[0127] Next, in step 1062, two-stage joint filtering is performed on the initial outliers. Spatial filtering is performed on the outliers, constructing a spherical neighborhood with a radius of R = 0.5 units around each point. The DBSCAN density clustering algorithm is used to detect spatial clustering. Only coordinates containing at least K outliers within the neighborhood are retained. Temporal filtering is performed on the outliers, counting their occurrence frequency within a sliding time window. A point must appear three or more times consecutively to be considered acceptable. A logical AND gate is used to combine the spatial clustering and temporal continuity conditions, outputting a set S of spatiotemporally correlated outliers. This step effectively eliminates isolated noise points, ensuring that the anomaly exhibits systematic characteristics.

[0128] Then, step 1063 is used to filter the outlier points after spatiotemporal filtering. Perform multi-dimensional feature extraction. First, calculate the spatial distribution range, traverse all point coordinates, obtain the maximum and minimum values ​​of the X / Y / Z axes respectively, and generate a three-dimensional bounding box. Then, the degree of spatial discreteness is quantified and the average Euclidean distance between all pairs of points in the set is calculated. This is achieved through a double loop. and Calculating distance , and finally calculate the arithmetic mean of the distances between all points , which reflects the diffusion characteristics of the fault signal. At the time series level, we construct a time feature sequence, divide the continuous time axis into 1-second unit windows, count the number of abnormal points in each window, and generate a frequency distribution histogram. , extract the highest frequency For example, 8Hz means 8 anomalies occur per second; arrange all points in timestamp order, calculate the time interval between adjacent points, and take the median of the interval sequence , for example, 0.2 seconds represents the typical interval of an abnormal event. Fusion of spatial and temporal features to construct a six-dimensional composite descriptor ,By matching the historical fault library samples by cosine similarity, when the maximum similarity exceeds 0.85, the corresponding fault type label such as “bearing crack” is assigned to the corresponding ,abnormal point.

[0129] Finally, through step 1064, the fault type label, spatial domain and initial timestamp are encapsulated into an industrial Internet of Things standard data packet and published through the MQTT protocol to form a traceable diagnosis result.

[0130] In practical applications, in the motor health monitoring of the six-axis drone pod, the vibration eigenvector 5.2mm / s, 128Hz, 0.87 is converted into a three-dimensional space coordinate point through t-SNE nonlinear mapping. The system calculates the Mahalanobis distance between this point and the historical fault cluster center ,because The outlier point is marked and the timestamp 2023-11-05T08:12:37.422Z is ​​recorded. Double filtering is then performed on the point. In the spatial dimension, a spherical neighborhood is constructed with a radius of R = 0.5 units, and 6 outliers are detected in the neighborhood. In the temporal dimension, 4 consecutive occurrences are counted within a 3-second sliding window, meeting the frequency N = 3, and thus included in the spatiotemporal associated outlier point set S. Multimodal feature fusion is performed on the set S, and the coordinate range is calculated in the spatial dimension. Characterize the fault spread range and spatial dispersion Reflects abnormal aggregation; the peak frequency of the unit time window extracted in the time dimension is 8Hz, representing 8 pulses per second, and the median of the interval between adjacent events The periodic pattern is revealed in seconds. The composite feature descriptor F=[0.7, 0.9, 1.1, 0.25, 8, 0.2] is generated by fusion, and the historical fault library is matched by cosine similarity to assign fault type labels. Finally, the diagnosis results are encapsulated as a list containing the fault type "winding short circuit", spatial domain The JSON-LD protocol data packet with the initial timestamp is published to the cloud platform via the MQTT protocol, and is located as an interphase short circuit fault caused by deterioration of the motor coil insulation, providing accurate traceability basis for operation and maintenance.

[0131] The overall solution of 106 mentioned above achieves accurate identification and tracing of equipment faults by constructing a multi-dimensional feature space mapping and dynamic anomaly detection mechanism. Its core value lies in the use of nonlinear coordinate transformation to quantify abstract features into measurable spatial locations, combined with an adaptive threshold algorithm to effectively distinguish normal fluctuations from potential anomalies. The innovative dual filtering mechanism integrates spatial aggregation and temporal continuity verification to systematically eliminate random noise interference. The fault matching technology based on composite feature descriptors deeply analyzes the essential characteristics of faults by integrating spatial distribution patterns and temporal evolution laws. The final output of standardized diagnostic results includes fault type, spatial domain location, and temporal traceability information, providing an actionable and accurate maintenance basis for industrial systems. The entire process forms a closed-loop technology chain from anomaly perception to fault diagnosis, significantly improving the reliability management level and predictive maintenance capabilities of complex equipment.

[0132] The following is a complete example for steps 101 to 106. Figure 2As shown, during the debugging of the six-axis logistics drone intelligent pod, a 200MHz bandwidth differential probe was used to non-invasively couple the 48V DC bus power line to capture the voltage fluctuation signal during the rapid rotation of the gimbal motor, detecting a periodic drop of 300mV / 20ms. A 50MHz magnetically isolated current sensor was simultaneously connected in parallel with the brushless motor control line to capture the 32A / 80ns transient spike current at the moment of phase switching, with a rising edge slope of 400A / μs. The two signals were synchronized with a 100MHz clock source to ensure that the timing deviation was less than 1ns. The voltage fluctuation signal was decomposed into three-band wavelet packets to extract the root mean square value of the low-frequency band of 0-1kHz, the peak value of the mid-frequency band of 1k-100kHz, and the energy proportion of the high-frequency band of 100k-10MHz to form the voltage segment characteristics, and the maximum drop amplitude of 300mV was recorded; the Teager-Kaiser energy operator was applied to the transient signal, and the integral value of the energy density within the 10ns window of the pulse leading edge was calculated to be 22.4mJ / Ω as the transient peak value, and the peak current of 32A and the single pulse energy of 4.7J were recorded simultaneously.

[0133] Next, hardware-level signal synchronization is established based on the PTP protocol. The synchronization trigger point is locked between the voltage fluctuation starting point t=2023-11-05T14:23:07.332Z and the transient pulse rising edge t=2023-11-05T14:23:07.3328Z. The ratio coefficient k=0.47V / A of the voltage drop of 0.3V and the transient amplitude of 32A is calculated, and the time difference is quantified. The time difference exceeds the normal threshold of 0.3ms, revealing the power supply response lag phenomenon. Based on the historical fault library statistics, the composite rule is generated, and the amplitude threshold interval is set to low frequency voltage, low frequency voltage , high-frequency energy ratio threshold> 8%; phase correlation value requires voltage transient time difference The time interval is less than or equal to 0.5ms and the cosine similarity is greater than 0.9. The composite rule integrates the coupling relationship between the amplitude parameter and the energy parameter.

[0134] Finally, a three-dimensional fault feature space is constructed, where the X-axis maps the voltage low-frequency stability, the Y-axis maps the transient energy density, and the Z-axis maps the time difference. A Gaussian mixture model was used to delineate the fault region: when X < 0.75 and Y > 0.3, it was designated as the "capacitor aging" region, and when Z > 0.8, it was designated as the "PWM interference" region. A dynamically updateable fault type prediction model was established. The feature vector [0.82, -1.15, 0.63] was mapped to the spatial point P (0.28, -0.92, 0.71) using t-SNE. Its Mahalanobis distance to the "capacitor aging" cluster center was calculated to be 4.3 (dynamic threshold θ = 3.9). Double filtering revealed seven outliers within a 0.5-unit radius neighborhood, with five consecutive triggering events within three seconds, exceeding three, generating a set of outlier points, S. The composite features of S, including spatial range, dispersion, main frequency, and interval median, were extracted. The fused descriptor F = [0.65, 0.81, 0.93, 0.28, 6, 0.18] had a similarity of 0.93 with the "capacitor aging" sample. After assigning a fault label, the diagnostic result was packaged and pushed to the operation and maintenance system via the MQTT protocol. Actual disassembly and verification showed that the ESR of the filter capacitor increased from 18 mΩ to 52 mΩ.

[0135] Figure 3 The present invention provides a schematic diagram of a structure of an intelligent fault diagnosis system for an intelligent pod, as shown in FIG. Figure 3 As shown, the system includes:

[0136] An acquisition module 31 is used to acquire a voltage fluctuation signal of the smart pod power line and a transient signal of the motor control line;

[0137] The processing module 32 is configured to perform multi-band processing on the voltage fluctuation signal to generate voltage segment features and record the amplitude parameters of the voltage fluctuation signal; perform feature extraction on the transient signal to generate transient peaks by capturing energy density changes in regions where the signal amplitude changes suddenly, and record the energy parameters of the transient signal;

[0138] An analysis module 33 is configured to establish a signal trigger point synchronization mechanism based on the voltage segment characteristics and the transient peak value, and perform a comparative analysis of the voltage fluctuation amplitude and the transient amplitude based on the synchronization trigger point, and calculate the time difference between the two;

[0139] A generating module 34 is configured to generate a composite rule including an amplitude threshold interval and a phase correlation value based on a preset periodic reference signal in combination with the amplitude parameter and the energy parameter;

[0140] An establishing module 35 is used to construct a multi-dimensional space including the amplitude parameter, the energy parameter and the time difference, and to establish a fault type prediction model in the multi-dimensional space based on the composite rule;

[0141] The prediction module 36 is used to identify abnormal points by calculating the matching degree between the distribution position of the feature vector in the multi-dimensional space and the fault type prediction model, and determine the corresponding fault type according to the distribution pattern of the abnormal points in the multi-dimensional space, and output a diagnostic result including the identified fault type and the corresponding abnormal point.

[0142] Figure 3 The intelligent fault diagnosis system for the intelligent pod can be executed Figure 1 The implementation principles and technical effects of the intelligent fault diagnosis method for a smart pod described in the illustrated embodiment are not further elaborated. The specific manner in which each module and unit performs operations in the intelligent fault diagnosis system for a smart pod in the above embodiment has been described in detail in the embodiments of the method and will not be further elaborated here.

[0143] In one possible design, Figure 3 The intelligent fault diagnosis system for an intelligent pod of the embodiment shown can be implemented as a computing device, such as Figure 4 As shown, the computing device may include a storage component 41 and a processing component 42;

[0144] The storage component 41 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 42 .

[0145] The processing component 42 is used for the above Figure 1 The embodiment provides an intelligent fault diagnosis method for an intelligent pod.

[0146] The processing component 42 may include one or more processors to execute computer instructions to perform all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0147] The storage component 41 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0148] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0149] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0150] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0151] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0152] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment provides an intelligent fault diagnosis method for an intelligent pod.

[0153] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0154] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0155] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent fault diagnosis method for an intelligent pod, characterized in that: include: Obtain voltage fluctuation signals of the smart pod power line and transient signals of the motor control line; Performing multi-band processing on the voltage fluctuation signal to generate voltage segment features, and recording the amplitude parameters of the voltage fluctuation signal; performing feature extraction on the transient signal to generate a transient peak by capturing energy density changes in the signal amplitude mutation region, and recording the energy parameters of the transient signal; Based on the voltage segment characteristics and the transient peak value, a signal trigger point synchronization mechanism is established, and based on the synchronization trigger point, a comparative analysis of the voltage fluctuation amplitude and the transient amplitude is performed to calculate the time difference between the two; Based on a preset periodic reference signal, in combination with the amplitude parameter and the energy parameter, a composite rule including an amplitude threshold interval and a phase correlation value is generated; Constructing a multidimensional space including the amplitude parameter, the energy parameter, and the time difference, and establishing a fault type prediction model in the multidimensional space based on the composite rule; By calculating the matching degree between the distribution position of the feature vector in the multi-dimensional space and the fault type prediction model, the abnormal points are identified, and the corresponding fault type is determined according to the distribution pattern of the abnormal points in the multi-dimensional space, and the diagnostic results including the identified fault type and the corresponding abnormal points are output.

2. The method according to claim 1, characterized in that The method includes calculating the matching degree between the distribution position of the feature vector in the multi-dimensional space and the fault type prediction model to identify abnormal points, determining the corresponding fault type according to the distribution pattern of the abnormal points in the multi-dimensional space, and outputting a diagnosis result including the identified fault type and the corresponding abnormal point, including: Converting the feature vector into a coordinate point in the multidimensional space through a preset nonlinear coordinate mapping function, calculating the geometric distance between the coordinate point and the historical benchmark data form in the fault type prediction model, and marking the corresponding coordinate point whose geometric distance exceeds a dynamic threshold as an abnormal point; Applying a dual filtering mechanism including spatial filtering and temporal filtering to the outliers to screen out outliers that meet both spatial aggregation and temporal continuity characteristics to generate an outlier set; Input the abnormal point set into the regional feature analysis module, extract the distribution geometric characteristics and time series characteristics of the abnormal point set, map the fault type label according to the distribution geometric characteristics and time series characteristics, and assign the corresponding abnormal point; The fault type label and the timestamp and spatial coordinates of the corresponding abnormal point are encapsulated into a diagnostic data packet that complies with the industrial Internet of Things protocol to output the diagnostic result.

3. The method according to claim 2, characterized in that The step of inputting the abnormal point set into a regional feature analysis module, extracting distribution geometric features and time series features of the abnormal point set, mapping fault type labels according to the distribution geometric features and time series features, and assigning corresponding abnormal points includes: In the regional feature analysis module, the coordinate span of the outlier point set in three-dimensional space is counted to determine the spatial distribution range, and the average distance between adjacent points is calculated to evaluate the spatial dispersion of the outlier point set. The spatial distribution range and spatial dispersion are used as distribution geometric features; At the same time, the frequency distribution of the outliers in the outlier set within a unit time is counted, and a time interval sequence is generated according to the difference between the timestamps of adjacent outliers. The frequency distribution and the time interval sequence are combined as time series features; The spatial distribution feature and the time series feature are fused to form a composite feature descriptor, and the composite feature descriptor is matched with the historical fault samples in the fault type prediction model for similarity. The fault type label corresponding to the matched fault mode is assigned to the corresponding abnormal point through feature similarity calculation.

4. The method according to claim 1, wherein The constructing of a multidimensional space including the amplitude parameter, the energy parameter, and the time difference, and establishing a fault type prediction model in the multidimensional space based on the composite rule, includes: Arranging the amplitude parameter, the energy parameter, and the time difference in chronological order into three-dimensional data units containing a timestamp identifier, and constructing a multi-dimensional space; Based on the composite rule and in combination with geometric distribution characteristics of the three-dimensional data units in the multi-dimensional space, clustering regions satisfying the composite rule are divided in the multi-dimensional space; Performing statistical analysis on the clustering area, extracting the cluster center coordinates, cluster range boundary values, and density distribution characteristics of the coordinate points within the cluster to form cluster feature data; The cluster feature data is matched with typical fault features in a pre-stored fault type library to generate a fault type prediction model including cluster center coordinate weights, dynamic adjustment coefficients of cluster range boundaries, and density distribution matching thresholds.

5. The method according to claim 1, wherein The method of establishing a signal trigger point synchronization mechanism based on the voltage segment characteristics and the transient peak value, and performing a comparative analysis of the voltage fluctuation amplitude and the transient amplitude based on the synchronization trigger point, and calculating the time difference between the two, includes: generating a dual constraint condition based on the amplitude duration and the amplitude change slope, screening the voltage abnormal signal segment whose amplitude parameter exceeds a first threshold in the voltage segment feature based on the dual constraint condition, and locating the energy mutation region whose energy parameter mutation exceeds a second threshold in the transient peak based on the consistency of timestamp alignment and energy density gradient direction; Establish a signal trigger point synchronization mechanism to compare the time overlap of the voltage abnormal signal segment and the energy mutation region point by point, and select matching events with timestamp differences less than a preset tolerance to mark them as synchronization trigger points; Taking the synchronous trigger point as a reference, the amplitude change slope of the voltage fluctuation signal before and after the trigger is calculated, and the energy density change rate of the transient peak before and after the trigger is calculated at the same time. The time difference between the voltage fluctuation amplitude and the starting moment of the transient amplitude is determined according to the intersection of the amplitude change slope and the energy density change rate.

6. The method according to claim 1, wherein The method of generating a composite rule including an amplitude threshold interval and a phase correlation value based on a preset periodic reference signal and combining the amplitude parameter and the energy parameter includes: Synchronously dividing a preset periodic reference signal into a plurality of periodic units of equal length according to the power frequency period of the power line; In each of the periodic units, an allowable fluctuation range of the amplitude parameter is set based on the statistical distribution of historical normal operating condition data to generate an amplitude threshold interval; Determining a phase correlation value between the energy parameter and the start time of the periodic unit according to the distribution density of the energy parameter; The amplitude threshold interval and the phase correlation value are logically combined to generate a composite rule including multiple logical operators and time window constraints.

7. The method according to claim 1, characterized in that The step of performing multi-band processing on the voltage fluctuation signal to generate voltage segment features and recording the amplitude parameters of the voltage fluctuation signal; performing feature extraction on the transient signal to generate a transient peak by capturing energy density changes in a region where the signal amplitude changes suddenly, and recording the energy parameters of the transient signal includes: The voltage fluctuation signal is divided into multiple basic units according to a preset time length. The frequency range of each basic unit is divided based on the fundamental frequency and harmonic frequency distribution range of the power line. The energy distribution value of the signal in each frequency range is counted and accumulated to form the voltage segment feature. detecting a maximum value and a minimum value of a signal amplitude in each basic unit according to the voltage segment characteristics, and taking a difference between the maximum value and the minimum value as an amplitude parameter corresponding to the basic unit; A sliding window is used to intercept the area in the transient signal where the amplitude change rate exceeds a preset condition as the amplitude mutation area, and the point of amplitude mutation in the amplitude mutation area is identified as the critical point. The energy density value corresponding to the critical point is taken as the transient peak value, and the total energy of the amplitude mutation area is accumulated to generate the energy parameter.

8. An intelligent fault diagnosis system for an intelligent pod, characterized in that: include: An acquisition module is used to obtain voltage fluctuation signals of the smart pod power line and transient signals of the motor control line; a processing module, configured to perform multi-band processing on the voltage fluctuation signal, generate voltage segment features, and record amplitude parameters of the voltage fluctuation signal; perform feature extraction on the transient signal, generate transient peaks by capturing energy density changes in regions where the signal amplitude changes suddenly, and record energy parameters of the transient signal; An analysis module is configured to establish a signal trigger point synchronization mechanism based on the voltage segment characteristics and the transient peak value, and to perform a comparative analysis of the voltage fluctuation amplitude and the transient amplitude based on the synchronization trigger point, and calculate the time difference between the two; A generating module, configured to generate a composite rule including an amplitude threshold interval and a phase correlation value based on a preset periodic reference signal in combination with the amplitude parameter and the energy parameter; An establishment module is used to construct a multi-dimensional space including the amplitude parameter, the energy parameter and the time difference, and establish a fault type prediction model in the multi-dimensional space based on the composite rule; The prediction module is used to identify abnormal points by calculating the matching degree between the distribution position of the feature vector in the multi-dimensional space and the fault type prediction model, and determine the corresponding fault type according to the distribution pattern of the abnormal points in the multi-dimensional space, and output a diagnostic result including the identified fault type and the corresponding abnormal point.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an intelligent fault diagnosis method for an intelligent pod as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, an intelligent fault diagnosis method for an intelligent pod according to any one of claims 1 to 7 is implemented.

Images