A method and system for diagnosing faults in electric actuators

Abstract

This invention discloses a method and system for fault diagnosis of electric actuators, belonging to the field of fault diagnosis technology. The method specifically includes: collecting current, voltage, displacement, temperature rise, and electromagnetic noise signals of the electric actuator during operation to form a multi-dimensional operating dataset; obtaining feature vectors by time-frequency domain fusion processing of the multi-dimensional operating data; constructing a health state trajectory of the electric actuator; introducing an electromagnetic noise coupling factor into the health state trajectory; performing evolutionary analysis on the health state trajectory; extracting trend information related to faults; and when the trend information exceeds a set threshold, determining the fault type and severity by comparing it with pre-stored actuator fault evolution samples. This application can achieve early identification and classification of potential faults through trend extraction and adaptive comparison, thereby improving the comprehensiveness and accuracy of diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, specifically a method and system for diagnosing faults in electric actuators. Background Technology

[0002] Electric actuators are widely used in aerospace, rail transportation, energy equipment, and automated production lines. Their core function is to convert electrical signals into mechanical displacement or torque to achieve precise execution. With the increasing complexity of application scenarios, electric actuators often operate in adverse conditions such as high temperatures, strong electromagnetic interference, and vibration and shock. This makes their internal drive motors, reduction mechanisms, and sensing components prone to performance degradation and potential failures. Traditional fault diagnosis methods are mostly based on the analysis of single current or voltage signals, lacking a comprehensive utilization of multi-dimensional operational characteristics.

[0003] In recent years, academia and industry have proposed various fault detection methods for electric actuators based on signal processing and pattern recognition. For example, frequency domain analysis is used to identify short circuits in motor windings, or temperature rise characteristics are used to determine transmission system jamming. However, these methods are usually limited to specific fault modes and often require preset static thresholds for judgment, lacking the ability to characterize the dynamic evolution of the health state. When the actuator is in the early stage of degradation, these methods are not sensitive enough and are prone to missed or false detections, failing to meet the application requirements of high reliability scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a fault diagnosis method and system for electric actuators, which can take into account multi-dimensional feature mapping, time evolution analysis and adaptive diagnosis, thereby improving the safety and maintainability of actuators under complex working conditions.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for diagnosing faults in an electric actuator, comprising:

[0007] Collect current, voltage, displacement, temperature rise and electromagnetic noise signals of electric actuators during operation to form a multi-dimensional operation dataset;

[0008] The multi-dimensional operating data is processed by time-frequency domain fusion to obtain feature vectors, and the health status trajectory of the electric actuator is constructed.

[0009] An electromagnetic noise coupling factor is introduced into the health state trajectory to perform evolution analysis on the health state trajectory and extract trend information related to the fault.

[0010] When the trend information exceeds a set threshold, the fault type and severity are determined by comparing it with pre-stored actuator fault evolution samples.

[0011] Specifically, the multi-dimensional operational data is processed through time-frequency domain fusion to obtain feature vectors, and a health status trajectory of the electric actuator is constructed, including:

[0012] The multi-dimensional operational data is mapped to the time domain, frequency domain, and time-frequency joint domain respectively to form three types of basic signal representations;

[0013] The three types of basic signal representations are respectively feature-encoded, and the correlation weight between actuator drive current and electromagnetic noise is introduced in the encoding process;

[0014] The feature encoding results are fused by sequence concatenation to obtain a composite feature matrix containing dynamic and static information.

[0015] In the composite feature matrix, the rearranged and refined state vector group is based on the sliding time slice;

[0016] The state vector groups are arranged sequentially to generate the health status trajectory of the electric actuator.

[0017] Specifically, the feature encoding results are fused using a sequence concatenation method to obtain a composite feature matrix containing dynamic and static information, including:

[0018] Each feature encoding result is sequentially indexed according to its source in the time domain, frequency domain, and time-frequency joint domain to form an initial feature sequence with domain labels;

[0019] In the initial feature sequence, the sequence is rearranged according to the alternation pattern of time domain, frequency domain, and time-frequency joint domain to obtain a multi-level interwoven arrangement sequence;

[0020] The permutation sequence is cyclically concatenated along the time dimension;

[0021] The permutation sequence after cyclic splicing is reconstructed into a two-dimensional matrix, which is a composite feature matrix containing dynamic and static features.

[0022] Specifically, in the composite feature matrix, the rearranged and refined state vector group based on the sliding time slice includes:

[0023] A sliding time slice sequence is generated by setting a preset window length and step size along the time dimension of the composite feature matrix.

[0024] In each time slice, a joint reference column containing current and electromagnetic noise codes is selected as the anchoring reference. Based on the timing mark of the anchoring reference, the position order of the feature columns to be aligned in the time slice is rearranged to form an alignment sub-matrix.

[0025] The alignment submatrix is ​​hierarchically split according to dynamic and static labels. First, deinterleaving is performed in the column dimension, and then convergence is performed in the row dimension to obtain the candidate vector group for the time slice.

[0026] The candidate vector groups are connected in a preset order to generate state vectors corresponding to the time slices, and then arranged in the order of the time slice index identifiers to form the state vector groups.

[0027] Specifically, the state vector groups are arranged sequentially to generate the health state trajectory of the electric actuator, including:

[0028] Based on the index identifiers of the state vectors in different time slices in the state vector group, the time order of each state vector is aligned.

[0029] The aligned state vectors are concatenated end-to-end according to time order to form a continuous vector sequence.

[0030] The continuous vector sequence is path-mapped in a multidimensional coordinate system;

[0031] Based on the path mapping result, the continuous vector sequence is organized into a single trajectory, which serves as the health status trajectory of the electric actuator.

[0032] Specifically, an electromagnetic noise coupling factor is introduced into the health state trajectory to perform evolution analysis on the health state trajectory and extract trend information related to the fault, including:

[0033] Based on the operating signal of the electric actuator, electromagnetic noise and current are jointly calibrated to generate an electromagnetic noise coupling factor.

[0034] The electromagnetic noise coupling factor is embedded as an additional dimension into the health state trajectory to form an extended trajectory with coupling characteristics.

[0035] The extended trajectory is fragmented according to a preset time window to generate trajectory segments;

[0036] Between the trajectory segments, a comparison operation is performed based on the change pattern of the electromagnetic noise coupling factor to filter out abnormal offset segments.

[0037] The abnormal offset segments are arranged and merged in chronological order to extract trend information related to potential faults.

[0038] Specifically, between the trajectory segments, a comparison operation is performed based on the variation pattern of the electromagnetic noise coupling factor to filter out abnormal offset segments, including:

[0039] The variation patterns of electromagnetic noise coupling factors are extracted from each trajectory segment to form corresponding pattern sequences;

[0040] The pattern sequences are matched one by one according to the time index, and the change patterns between different trajectory segments are compared and analyzed.

[0041] In the comparative analysis, the locations of segments with significantly different change patterns were identified, forming a set of candidate abnormal segments;

[0042] The candidate abnormal segment set is compared a second time to filter out trajectory segments whose deviation exceeds a set threshold and these segments are identified as abnormal deviation segments.

[0043] Specifically, the abnormal offset segments are arranged and merged in chronological order to extract trend information related to potential faults, including:

[0044] Based on the time index of the abnormal offset segment, the segments are arranged in chronological order to form an initial sequence;

[0045] In the initial sequence, adjacent time segments are spliced ​​together continuously to eliminate time gaps and form a locally extended sequence;

[0046] Cross-segment fusion is performed on the local extended sequence to integrate abnormal fragments distributed in different time periods and generate a global sequence;

[0047] The global sequence is expanded along the time axis and reconstructed into a single time series, which serves as trend information related to potential faults.

[0048] Specifically, when the trend information exceeds a set threshold, the fault type and severity are determined by comparing it with pre-stored actuator fault evolution samples, including:

[0049] The system detects the value of trend information, and when it exceeds a preset threshold, it initiates a comparison process.

[0050] Retrieve a set of actuator failure evolution samples corresponding to the current operating conditions from the pre-stored database;

[0051] A one-to-one mapping relationship is established between the trend information and the key features of the fault evolution samples to form a comparison matrix;

[0052] In the comparison matrix, the trend information and the fault evolution samples are compared segment by segment based on the time evolution order to select the candidate samples with the highest matching degree.

[0053] Based on the annotation information of the candidate samples, the fault type and severity of the electric actuator are determined.

[0054] An electric actuator fault diagnosis system is provided to implement the aforementioned electric actuator fault diagnosis method, comprising: a data acquisition module, a trajectory construction module, a fault information extraction module, and a discrimination module;

[0055] The data acquisition module is used to collect current, voltage, displacement, temperature rise and electromagnetic noise signals of the electric actuator during operation, forming a multi-dimensional operation dataset;

[0056] The trajectory construction module is used to obtain feature vectors by time-frequency domain fusion processing of the multi-dimensional operating data, and to construct the health status trajectory of the electric actuator.

[0057] The fault information extraction module is used to introduce an electromagnetic noise coupling factor into the health state trajectory, perform evolution analysis on the health state trajectory, and extract trend information related to the fault.

[0058] The discrimination module is used to determine the fault type and severity by comparing the trend information with pre-stored actuator fault evolution samples when the trend information exceeds a set threshold. Compared with the prior art, the beneficial effects of the present invention are:

[0059] This invention proposes a fault diagnosis method and system for electric actuators. It constructs a health state trajectory and introduces an electromagnetic noise coupling factor into the trajectory for evolution analysis. This trajectory is then compared and judged against pre-stored fault evolution samples, thereby achieving a full-process diagnosis of the electric actuator's operating status. This overall method not only presents the dynamic evolution process of the actuator in a unified health space but also achieves early identification and classification of potential faults through trend extraction and adaptive comparison, thereby improving the comprehensiveness and accuracy of the diagnosis and enhancing the operational reliability and maintainability of electric actuators under complex working conditions. Attached Figure Description

[0060] Figure 1 A flowchart of an electric actuator fault diagnosis method provided by the present invention;

[0061] Figure 2 A structural schematic diagram of the health status trajectory of the electric actuator provided by the present invention;

[0062] Figure 3 This invention provides an architecture diagram for an electric actuator fault diagnosis system. Detailed Implementation

[0063] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," and "third" used in this application do not limit the data or execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0066] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0067] Example 1

[0068] Please see Figures 1-2 The present invention provides an embodiment of a method comprising the following specific steps:

[0069] Step S1: Collect current, voltage, displacement, temperature rise and electromagnetic noise signals of the electric actuator during operation to form a multi-dimensional operation dataset.

[0070] In this embodiment, based on the synchronous acquisition and normalization processing of multi-source physical signals, specifically, current and voltage signals are acquired through detection modules installed at the drive circuit nodes to characterize the electromagnetic drive state of the actuator motor; displacement signals are acquired through position sensors to reflect the motion behavior of the output end; temperature rise signals are obtained through temperature acquisition units deployed on the windings or housing to indicate changes in thermal characteristics; and electromagnetic noise signals are extracted through a high-frequency pickup module to capture implicit anomalies caused by electromagnetic field disturbances. It should be noted that the acquired multi-source signals are first aligned with a unified time reference to eliminate differences in sampling frequencies and phases of different sensors, and then amplitude normalization is performed to ensure the comparability of each signal in the same data space. Through the above steps, the resulting multi-dimensional operating dataset can completely reflect the dynamic state of the electric actuator under electrical, mechanical, thermal, and electromagnetic environments.

[0071] Step S2: The multi-dimensional operating data is processed by time-frequency domain fusion to obtain feature vectors, and the health status trajectory of the electric actuator is constructed.

[0072] like Figure 2 As shown, the specific steps of step S2 are as follows:

[0073] Step S201: Map the multi-dimensional running data to the time domain, frequency domain, and time-frequency joint domain respectively to form three types of basic signal representations.

[0074] In this embodiment, multi-domain signal reconstruction is performed on the collected multi-dimensional operational data. Specifically, firstly, low-frequency dynamic features such as current, voltage, and displacement are directly retained in the time domain to characterize the original response features of the actuator over time. Secondly, the temperature rise signal and electromagnetic noise signal are mapped to the frequency domain through frequency decomposition processing to separate periodic disturbances and high-frequency harmonic components, thereby revealing the energy distribution and resonance characteristics during operation. Furthermore, based on the frequency domain analysis, a sliding time window and energy distribution rearrangement strategy are introduced to synchronously superimpose time-domain and frequency-domain information, thus obtaining a time-frequency joint domain representation. It should be noted that after the above steps, the multi-dimensional operational data is divided into three basic signal representations: time-domain signals that reflect dynamic trends, frequency-domain signals that reflect periodic patterns, and time-frequency joint signals that can simultaneously capture local mutations and periodic features.

[0075] Step S202: Perform feature encoding on the three types of basic signal representations respectively, and introduce the correlation weight between actuator drive current and electromagnetic noise during the encoding process.

[0076] In this embodiment, digital encoding is performed based on the differentiated characteristics of three types of basic signal representations. Specifically, firstly, the amplitude distribution and dynamic change trend of the time-domain signal are extracted through statistical time-series features and converted into a vector representation. Secondly, the frequency-domain signal is extracted by summarizing energy distribution and harmonic components to form a spectral feature vector, which is used to preserve periodic patterns. Thirdly, the time-frequency joint signal is used to establish a time-frequency feature vector through window decomposition and local energy trajectory extraction to characterize transient disturbances and local burst characteristics. It should be noted that a correlation weight between actuator drive current and electromagnetic noise is introduced in the feature encoding process of various signals. That is, the coordinated change of the two is used as an enhancement factor when allocating encoding weights, so that the encoded result can highlight the electromechanical coupling effect. Through the above steps, the resulting set of weighted feature vectors not only distinguishes the feature attributes of different signal domains, but also preserves the correlation between current and electromagnetic noise at the encoding level.

[0077] Step S203: The feature encoding results are fused by sequence concatenation to obtain a composite feature matrix containing dynamic and static information.

[0078] The specific steps of step S203 are as follows:

[0079] Step S2031: Index each feature encoding result in sequence according to its source time domain, frequency domain and time-frequency joint domain to form an initial feature sequence with domain labels.

[0080] In this embodiment, based on the ordered organization of feature encoding results from different sources, specifically, firstly, the feature codes from the time domain are arranged according to the time evolution order to preserve the continuity of signal changes over time; secondly, the feature codes from the frequency domain are arranged in order from low to high frequency ranges to reflect the hierarchical relationship of energy distribution in different frequency bands; furthermore, the feature codes from the time-frequency joint domain are sorted according to the dual indexes of time window and frequency range to ensure that transient features and local harmonic components can be accurately located in the overall sequence. It should be noted that after completing the above sorting, a corresponding domain label is added to each encoding unit to indicate its signal source and feature category, thereby forming an initial feature sequence with clear domain identification. After this processing, the originally scattered feature vectors are organized into an ordered set with time, frequency, and time-frequency joint information.

[0081] Step S2032: In the initial feature sequence, rearrange it according to the alternation pattern of time domain, frequency domain and time-frequency joint domain to obtain a multi-level interwoven arrangement sequence.

[0082] In this embodiment, based on the serialization rearrangement strategy of the initial feature sequence, specifically, firstly, an alternation mode is set during the arrangement process, and time-domain features, frequency-domain features, and time-frequency joint-domain features are inserted alternately in a preset order to avoid the concentrated distribution of the same type of features in the sequence; secondly, during the alternating insertion process, by maintaining the consistency of the time index, it is ensured that the time-domain features can maintain a synchronous positional relationship with the frequency-domain features and time-frequency joint-domain features under their corresponding time slices; furthermore, a hierarchical control rule is introduced when alternating cross-domain features, that is, the pairing relationship between the time domain and the frequency domain is maintained first within a longer time slice, and the time-frequency joint-domain features are then embedded within a shorter time slice, forming a multi-level interleaved structure; it should be noted that after the above steps, the resulting arrangement sequence reflects a balanced distribution across domains macroscopically, and has a dual correspondence between time and frequency microscopically, thus forming a multi-level interleaved arrangement sequence.

[0083] Step S2033: The permutation sequence is cyclically spliced ​​in the time dimension.

[0084] In this embodiment, based on the temporal extension and reconstruction of the aforementioned multi-layered interwoven permutation sequence, specifically, firstly, using time slices as the basic unit, the permutation sequences corresponding to adjacent time slices are joined end-to-end to ensure continuity across segments; secondly, after completing the basic splicing, the spliced ​​sequence is cyclically extended, that is, the end segments are remapped to the starting position of the sequence according to the time index, and a periodic loop connection relationship is constructed in the sequence to form a loop structure; furthermore, the consistency of domain labels is maintained during the cyclic splicing process, so that the positions of the same type of feature in different time slices can form a traceable sequence channel through cyclic mapping; it should be noted that the cyclic spliced ​​sequence obtained by the above operations not only has extensibility in the temporal dimension, but also exhibits periodicity and traceability in the logical structure.

[0085] Step S2034: Reconstruct the cyclically concatenated permutation sequence into a two-dimensional matrix, wherein the two-dimensional matrix is ​​a composite feature matrix containing dynamic and static features.

[0086] In this embodiment, based on the structured transformation of the cyclically spliced ​​sequence, specifically, firstly, the spliced ​​sequence elements are mapped one by one onto a two-dimensional coordinate plane, with the time index as the row dimension and the feature domain label as the column dimension; secondly, the temporal order is maintained in the row dimension arrangement, so that features at different times can be unfolded sequentially along the vertical direction; furthermore, the alternating order of the time domain, frequency domain, and time-frequency joint domain is maintained in the column dimension distribution, so that features from different sources form a parallel correspondence in the horizontal direction; finally, in the matrix construction process, the dynamic changes between time slices and the static distribution of cross-domain features are preserved, so as to achieve the synchronous integration of dynamic and static information. It should be noted that the two-dimensional matrix reconstructed by this operation has the consistency of multi-dimensional signals in both the temporal and feature domain dimensions, forming a composite feature matrix that can simultaneously carry dynamic evolution features and static steady-state features.

[0087] Step S204: In the composite feature matrix, rearrange and refine the state vector group based on the sliding time slice.

[0088] The specific steps of step S204 are as follows:

[0089] Step S2041: Set the preset window length and step size along the time dimension of the composite feature matrix to generate a sliding time slice sequence.

[0090] In this embodiment, based on segmented sampling of the time dimension of the composite feature matrix, specifically, firstly, a fixed-length window is set on the vertical time axis of the matrix, which serves as the boundary range of a time slice; secondly, the step size of the window is determined so that the window slides on the time axis at preset intervals, generating a new time slice interval with each slide; furthermore, partial overlap between time slices is allowed during the sliding process to ensure that the continuity across segments is not disrupted, while allowing adjacent time slices to retain shared temporal information; finally, by continuously dividing the entire time dimension in the above manner, a sliding time slice sequence covering the complete running cycle is obtained. It should be noted that the sliding time slice sequence formed by this processing decomposes the composite feature matrix into a series of time slices, each of which retains the local representation of multi-domain features within that time range.

[0091] Step S2042: In each time slice, select a joint reference column containing current and electromagnetic noise codes as the anchoring reference. Based on the timing mark of the anchoring reference, rearrange the position order of the feature columns to be aligned in the time slice to form an alignment sub-matrix.

[0092] In this embodiment, based on establishing a unified alignment reference in the sliding time slice, specifically, firstly, a joint reference column containing both current signal encoding and electromagnetic noise encoding is identified in the feature columns of each time slice, and this is used as an anchoring reference; secondly, the timing mark information in the joint reference column is extracted, which serves as the alignment axis within the time slice; furthermore, the remaining feature columns to be aligned within the time slice are rearranged according to the timing mark, so that they correspond to the reference column in the vertical position; finally, the rearranged result is integrated into an alignment sub-matrix, in which all feature columns are consistent with the reference column in the time dimension. It should be noted that, through the above operations, the multi-source features in each time slice can be reorganized under the same reference, eliminating the offset problem caused by sampling frequency differences or signal drift.

[0093] Step S2043: The aligned sub-matrix is ​​hierarchically split according to dynamic and static labels. First, deinterleaving is performed in the column dimension, and then convergence is performed in the row dimension to obtain the candidate vector group for the time slice.

[0094] In this embodiment, based on the multi-level structured splitting of the aligned submatrix, specifically, firstly, according to the pre-defined feature attribute identifiers, the feature columns in the matrix are divided into dynamic marker columns with time-varying characteristics and static marker columns reflecting steady-state characteristics; secondly, the dynamic and static marker columns are de-interleaved in the column dimension, that is, the originally interleaved feature columns are regrouped so that similar feature columns are aggregated in the matrix; thirdly, after completing the column dimension grouping, the feature rows within the same group are converged in the row dimension. Through the regularization of the time-series index and the integration of redundant information, multiple rows of data are compressed into representative vector units; finally, through the dual processing of the column and row dimensions, each time slice is converted into a set of candidate vectors, corresponding to the simplified expressions of dynamic and static features respectively. It should be noted that this processing method not only enables the candidate vector group to maintain the hierarchical independence of different feature categories, but also ensures its temporal integrity.

[0095] Step S2044: Connect the candidate vector groups in a preset order to generate state vectors for the corresponding time slices, and arrange them in the order of the time slice index identifiers to form the state vector groups.

[0096] In this embodiment, based on the sequential integration of candidate vectors and the global arrangement of time-slice indices, specifically, firstly, the candidate vector groups formed within each time slice are connected according to a preset splicing order, which is led by dynamic feature vectors and supplemented by static feature vectors, ensuring that features of different attributes have a fixed positional relationship in the same state vector; secondly, after completing the intra-slice connection, a state vector is generated for each time slice, which serves as the overall representation of the running state of that segment; furthermore, according to the time-slice index identifier, the state vectors generated by different time slices are sorted and arranged according to the time sequence, so that the state vectors maintain strict temporal continuity; finally, the state vectors of all time slices are sequentially connected to form a state vector group covering the entire running cycle. It should be noted that the state vector group formed by the above operations not only retains the local features of each time slice, but also achieves continuous expression across segments through global time indexing.

[0097] Step S205: Arrange the state vector groups sequentially to generate the health state trajectory of the electric actuator.

[0098] The specific steps of step S205 are as follows:

[0099] Step S2051: Align the state vectors in time order according to their index identifiers in different time slices.

[0100] In this embodiment, based on the unified sorting of state vectors on the global time axis, specifically, firstly, the time slice index identifier corresponding to each state vector is read, which serves as the unique reference for the time order; secondly, all state vectors are compared according to their index identifiers, and duplicate or missing index positions are filled or removed to ensure the continuity and integrity of the index sequence; furthermore, the internal structure of the state vectors remains unchanged during the alignment process, only their positions in the sequence are adjusted, thereby avoiding the destruction of feature representation; finally, through the above processing, the state vectors of different time slices are rearranged into an arrangement that strictly follows the time order; it should be noted that this operation enables the state vector group to form a unified alignment benchmark in the time dimension.

[0101] Step S2052: Concatenate the aligned state vectors end-to-end according to the time sequence to form a continuous vector sequence.

[0102] In this embodiment, based on the temporal concatenation of aligned state vectors, specifically, firstly, the aligned state vectors are read one by one in time index order and arranged sequentially in the sequence according to the time slice order; secondly, during the arrangement process, the state vectors of adjacent time slices are directly concatenated by joining the first and last ends, so that there are no gaps or jumps between time slices; furthermore, to ensure the continuity of the sequence, the complete feature dimension of each state vector is preserved during concatenation, and a connection marker is established at the boundary position to ensure that the features of different time slices can maintain a smooth transition logically; finally, all state vectors are organized into a continuous vector sequence covering the entire time domain through the concatenation operation of joining the first and last ends. It should be noted that this operation enables the seamless connection of local time slice features on the global time axis.

[0103] Step S2053: Path-map the continuous vector sequence in a multidimensional coordinate system.

[0104] In this embodiment, based on the spatial representation of a continuous vector sequence, specifically, firstly, each state vector in the sequence is assigned a corresponding coordinate dimension. Dynamic features are mapped to the principal axis dimension to reflect temporal evolution, static features are mapped to the lateral dimension to reflect steady-state distribution, and composite features are projected to the auxiliary dimension to preserve cross-domain information. Secondly, each state vector is projected point by point into the multi-dimensional coordinate system in chronological order, forming sequentially connected nodes in space. Thirdly, connection paths are established between nodes according to their temporal order, thereby expanding the original one-dimensional sequence into a spatial path. Finally, the entire continuous vector sequence is transformed into a directional and hierarchical path representation in the multi-dimensional coordinate system. It should be noted that, through the above mapping, the sequence features not only maintain continuity on the time axis but also form a trajectory-like expression in multi-dimensional space.

[0105] Step S2054: Based on the path mapping result, organize the continuous vector sequence into a single trajectory and use it as the health status trajectory of the electric actuator.

[0106] In this embodiment, based on the global convergence of the path mapping results, specifically, firstly, the nodes connected sequentially in the multi-dimensional coordinate system are path-normalized to smooth out minor offsets caused by noise or local disturbances, maintaining the continuity of the overall path; secondly, based on the path normalization, the connection relationships of all nodes are converged into a single main trajectory to avoid multiple branches or repeated loops, ensuring that the trajectory has a unique direction in space; furthermore, after the single trajectory is formed, a clear time coordinate and feature attribution are established for each location point in the trajectory through dual annotation of time index and feature dimension, thereby ensuring the traceability and resolvability of the trajectory; finally, the processed continuous vector sequence is defined as the health status trajectory of the electric actuator, which logically runs through the entire operating cycle and is presented in space as a single trajectory; it should be noted that this processing realizes the transformation from a multi-dimensional complex sequence to a globally unique trajectory.

[0107] like Figure 2 As shown in the figure, a two-dimensional coordinate system is used as the mapping space, where the horizontal axis X represents the comprehensive evolution sequence of multi-dimensional features during operation, and the vertical axis Y represents the amplitude of state change at the same time scale. Through the aforementioned feature extraction and state vector construction steps, the original multi-source operation data is transformed into a continuous vector group, and after path mapping in the multi-dimensional coordinate system, it is reconstructed into a single trajectory. The spiral curve shown is the trajectory of this healthy state, and its extension from the inside to the outside corresponds to the entire process of the electric actuator's gradual evolution from the initial normal state to the operating state. It should be noted that this trajectory not only maintains the continuity of dynamic and static features in time, but also exhibits regular convergence and expansion in spatial distribution, which is convenient for identifying potential abnormal deviations and fault trends in the subsequent evolution analysis stage.

[0108] Step S3: Introduce an electromagnetic noise coupling factor into the health state trajectory, perform evolution analysis on the health state trajectory, and extract trend information related to the fault.

[0109] The specific steps of step S3 are as follows:

[0110] Step S301: Based on the operating signal of the electric actuator, the electromagnetic noise and current are jointly calibrated to generate an electromagnetic noise coupling factor.

[0111] In this embodiment, based on the correlation modeling of current and electromagnetic noise in the operating signal of the electric actuator, specifically, firstly, the drive current signal and external electromagnetic noise signal are synchronously acquired during the operation of the actuator, and the two types of signals are time-aligned to eliminate the deviation caused by sampling delay; secondly, by analyzing the correspondence between the amplitude change of the current signal and the spectral characteristics of the electromagnetic noise, the resonance interval and synchronous disturbance mode of the two within the same time slice are extracted; furthermore, this correspondence is normalized and mapped into a set of joint calibration parameters, so that the dynamic change of the current and the disturbance amplitude of the electromagnetic noise can be compared with the same dimension; finally, the above joint calibration parameters are integrated to form a quantifiable electromagnetic noise coupling factor, which is used to characterize the coupling strength between electromagnetic disturbance and current fluctuation under electromechanical action; it should be noted that, through this processing, the coupling relationship between electromagnetic noise and current is refined into a single index.

[0112] Step S302: Embed the electromagnetic noise coupling factor as an additional dimension into the health state trajectory to form an extended trajectory with coupling characteristics.

[0113] In this embodiment, based on the dimensional expansion processing of the health state trajectory, specifically, firstly, the position of the state vector corresponding to each time slice is determined in the generated health state trajectory, and the electromagnetic noise coupling factor is introduced as an independent additional quantity; secondly, a separate dimensional channel is established for this factor in each state vector, and it is stored in parallel with the original dynamic and static features to ensure that different categories of features have a fixed arrangement order within the vector; furthermore, the time index of the electromagnetic noise coupling factor is kept consistent with the original state vector in the overall trajectory structure, so that it can evolve synchronously with other features during path mapping; finally, after the above embedding processing, the trajectory, which was originally composed only of multi-domain features, is expanded into a trajectory with coupling features, so that the trajectory has an additional dimension in space that can directly reflect the coupling relationship between current and electromagnetic noise; it should be noted that this expanded trajectory introduces electromechanical coupling features while maintaining the continuity of the health state trajectory.

[0114] Step S303: The extended trajectory is fragmented according to a preset time window to generate trajectory segments.

[0115] In this embodiment, based on the time segmentation strategy of the extended trajectory, specifically, firstly, a preset time window length is set on the time axis after the health state trajectory is extended, and the global trajectory is divided into several continuous segments by sliding along the trajectory segment by segment with a fixed step size; secondly, the consistency of each feature dimension in the trajectory is maintained during the segmentation process, so that dynamic features, static features, and electromagnetic noise coupling factors can be retained simultaneously in the same segment; furthermore, the overlapping parts of the segments caused by the window boundary are uniformly marked, so that adjacent segments can retain shared information without causing data fragmentation when compared in subsequent times; finally, the set of trajectory segments obtained by segmentation not only covers the global operation process, but also presents the fine-grained changes of features over time in the local range; it should be noted that this segmentation process makes the extended trajectory segmented and comparable.

[0116] Step S304: Compare the trajectory segments according to the change pattern of the electromagnetic noise coupling factor to filter out abnormal offset segments.

[0117] The specific steps of step S304 are as follows:

[0118] Step S3041: Extract the variation patterns of electromagnetic noise coupling factors from each trajectory segment to form the corresponding pattern sequence.

[0119] In this embodiment, based on the temporal analysis of electromagnetic noise coupling factors in trajectory segments, specifically, firstly, the time series corresponding to the electromagnetic noise coupling factors is extracted within each trajectory segment, and segmented by amplitude fluctuations and directional changes to capture their basic characteristics of evolution over time; secondly, the continuous change process within the segment is summarized into several mode units, such as steady-state maintenance, sudden increases and decreases, or periodic oscillations, and a unique mode label is assigned to each unit; furthermore, these mode labels are arranged sequentially in chronological order within the same segment to form a mode sequence that reflects the evolution trend of coupling factors within the segment; finally, the mode sequences generated for each trajectory segment are stored independently for subsequent cross-segment comparative analysis and difference comparison; it should be noted that this processing method transforms complex numerical changes into structured mode sequences, enabling the temporal characteristics of the electromagnetic noise and current coupling relationship to be expressed in a symbolic form.

[0120] Step S3042: The pattern sequence is matched one by one according to the time index, and the change patterns between different trajectory segments are compared and analyzed.

[0121] In this embodiment, based on the index-by-index comparison strategy of cross-segment pattern sequences, specifically, firstly, time index identifiers are added to the pattern sequences generated by each trajectory segment to ensure that the positional correspondence of different segments on the global time axis is clear; secondly, during the comparison process, based on the time index, pattern units at the same index position are matched one-to-one, and their change trends are checked for matching to identify continuity and differences; furthermore, for pattern units that are inconsistent under adjacent time indices, their offset positions in the global trajectory are recorded by establishing difference markers; finally, through one-to-one correspondence and difference marker processing, a cross-segment pattern comparison result is obtained, which can reveal the continuity and imbalance of the electromagnetic noise coupling factor change pattern in different time periods; it should be noted that this method aligns and compares the local pattern sequences of each segment in the global dimension by introducing time index constraints.

[0122] Step S3043: Identify the locations of segments with significantly different change patterns in the comparative analysis to form a set of candidate abnormal segments.

[0123] In this embodiment, based on the difference localization in the comparison analysis results, specifically, firstly, in the one-to-one pattern sequence comparison results, index positions that are significantly inconsistent with the reference pattern are retrieved, and their corresponding trajectory segment identifiers are recorded; secondly, by aggregating and statistically analyzing these difference positions, segment regions that deviate multiple times within a continuous time interval are identified, enhancing the stability of difference detection; furthermore, based on the aggregation and statistics, segment regions with difference frequencies exceeding a preset threshold are marked as candidate anomalous segments, and their index positions and corresponding pattern features are retained in the set; finally, the resulting set of candidate anomalous segments serves as an anomalous candidate interval in the extended trajectory for further secondary screening and trend extraction. It should be noted that this process, through difference aggregation and threshold screening, transforms pattern inconsistencies into a traceable set of segments, achieving the initial capture of anomalous evolution regions.

[0124] Step S3044: Perform a second comparison on the candidate abnormal segment set, filter out trajectory segments whose offset exceeds a set threshold, and use them as abnormal offset segments.

[0125] In this embodiment, based on the refined screening of candidate abnormal segments, specifically, firstly, the change curve of the electromagnetic noise coupling factor is re-extracted for each trajectory segment in the candidate abnormal segment set, and the difference is calculated with its reference mode in the normal range to quantify the degree of segment offset; secondly, the obtained offset degree is compared with a preset threshold to preliminarily determine whether it meets the abnormal standard; furthermore, for segments close to the threshold boundary, the mode consistency check of adjacent time periods is further used to confirm whether it belongs to a real anomaly, thereby avoiding misjudgment caused by instantaneous fluctuations; finally, segments with offset degree exceeding the threshold and passing the consistency check are included in the abnormal offset segment set, and segments that do not meet the conditions are removed. It should be noted that this step, through the combination of secondary comparison and threshold control, achieves the convergence of screening from candidate abnormal segments to final abnormal offset segments.

[0126] Step S305: Arrange and merge the abnormal offset segments in chronological order to extract trend information related to potential faults.

[0127] The specific steps of step S305 are as follows:

[0128] Step S3051: Arrange each segment in chronological order according to the time index of the abnormal offset segment to form an initial sequence.

[0129] In this embodiment, based on the temporal organization of anomalous offset segments, specifically, firstly, the time index identifier corresponding to each anomalous offset segment is extracted and used as the sole basis for sorting; secondly, all segments in the set are arranged in ascending order according to the size of their time indices, so that the segments are sequentially connected on the global time axis; furthermore, for multiple segments appearing under the same time index, the segment with the strongest representativeness is selected and retained by comparing their feature tags to avoid interference from duplicate data in subsequent processing; finally, the initial sequence obtained after the above processing not only ensures the continuity of anomalous offset segments in the time dimension, but also provides logically ordered input for subsequent continuous splicing and trend extraction; it should be noted that this process normalizes the originally disordered distribution of anomalous segments into an initial sequence arranged in chronological order.

[0130] Step S3052: In the initial sequence, adjacent time segments are spliced ​​together continuously to eliminate time gaps and form a locally extended sequence.

[0131] In this embodiment, based on the continuous processing of the initial sequence, specifically, firstly, adjacent time segments in the initial sequence are searched one by one to confirm whether there are gaps in their time indices; secondly, for segments with continuous indices, their boundary positions are directly joined end to end, so that the feature vectors can form an uninterrupted connection in the time dimension; furthermore, for segments with small index intervals but not completely continuous, missing intervals are supplemented by interpolation or adjacent feature alignment, thereby eliminating time gaps caused by uneven sampling or segment segmentation; finally, the spliced ​​result forms an extended sequence in a local range, which can completely cover the continuous time interval of adjacent segments and retain the transition relationship of feature evolution; it should be noted that this step transforms the initial sequence from discrete segments into a continuous expression through local splicing and gap elimination.

[0132] Step S3053: Perform cross-segment fusion on the local extended sequence to integrate abnormal segments distributed in different time periods and generate a global sequence.

[0133] In this embodiment, based on the global fusion processing of local extended sequences, specifically, firstly, the start and end index intervals of the local extended sequences generated in different time periods are extracted to clarify the position range of each sequence on the global time axis; secondly, boundary comparisons are performed on these local sequences. For intervals with temporal overlap, a weighted merging method based on feature values ​​is used to achieve a unified expression of the overlapping segments, while for intervals with temporal gaps, interpolation or proximity patterns are used to fill in the missing content to ensure overall continuity; furthermore, consistency is maintained in the dimensions of dynamic features, static features, and coupling factors during the fusion process to avoid information misalignment when splicing across segments; finally, through cross-segment fusion processing, the local abnormal extended sequences distributed in different time periods are integrated into a continuous sequence covering the entire time axis. It should be noted that this global sequence not only retains the abnormal feature information in each local segment but also achieves overall coherence in the time dimension through a unified indexing system.

[0134] Step S3054: Expand the global sequence according to the time axis, reconstruct it into a single time series, and use it as trend information related to potential faults.

[0135] In this embodiment, the temporal reconstruction based on the global sequence involves several steps. First, the global sequence obtained through cross-segment fusion is expanded point-by-point according to the time index, allowing the abnormal features of each segment in the sequence to be continuously arranged on a unified time axis. Second, during the expansion process, the original dynamic features, static features, and electromagnetic noise coupling factors of each data point are preserved, and any duplicate nodes are deredundant to ensure the uniqueness and integrity of the sequence. Third, after the time axis expansion is completed, the global sequence, originally composed of multiple segments, is integrated into a single time series, allowing the abnormal features to be presented as a continuous curve. Finally, this single time series is defined as trend information related to potential faults, providing an intuitive and quantifiable input for subsequent diagnostic and judgment steps. It should be noted that this reconstruction process unifies the scattered abnormal features onto a continuous time line, realizing the transformation from local anomalies to global trends, laying the analytical foundation for fault type identification and severity determination.

[0136] Step S4: When the trend information exceeds the set threshold, the fault type and severity are determined by comparing it with the pre-stored actuator fault evolution samples.

[0137] The specific steps of step S4 are as follows:

[0138] Step S401: Detect the value of the trend information. When it exceeds the preset threshold, start the comparison process.

[0139] In this embodiment, the threshold triggering mechanism based on trend information specifically involves: first, extracting the numerical distribution of trend information point by point in a single time series and statistically analyzing its fluctuation range on the time axis; second, comparing the obtained trend values ​​with a pre-set threshold item by item, and determining that a trigger condition is met when the value within a continuous interval is consistently higher than the threshold; third, to avoid interference from instantaneous anomalies, a time window accumulation judgment is introduced during the comparison process, i.e., only when the proportion of values ​​continuously exceeding the threshold within the window reaches a set standard is it considered an over-limit event; finally, once an over-limit event is confirmed, the subsequent comparison process is immediately activated, entering the stage of fault type and severity determination; it should be noted that this detection method, through the combination of numerical comparison and time window accumulation, transforms abnormal fluctuations in trend information into stable trigger signals.

[0140] Step S402: Retrieve the set of actuator fault evolution samples corresponding to the current operating conditions from the pre-stored database.

[0141] In this embodiment, the retrieval is based on the matching of operating conditions and a sample library. Specifically, firstly, an operating condition index is generated according to the current operating status information of the electric actuator (including load level, ambient temperature, power supply characteristics, and motion mode); secondly, this index is compared with fault evolution samples collected and organized under different operating conditions in a pre-stored database, and a sample set that highly corresponds to the current operating conditions in terms of key characteristic parameters is selected; furthermore, samples with excessively different operating environments from the current situation are removed during the selection process to ensure the validity of the comparison data; finally, the obtained sample set covers the types of faults and evolution trajectories that may occur in the actuator under similar operating conditions, which can be directly used as a reference object for subsequent comparison and judgment. It should be noted that this step, through the bidirectional matching of operating conditions and sample characteristics, makes the retrieval of fault evolution samples targeted and applicable.

[0142] Step S403: Establish a one-to-one mapping relationship between the trend information and the key features of the fault evolution sample to form a comparison matrix.

[0143] In this embodiment, modeling is based on the correspondence between trend information and fault evolution sample features. Specifically, firstly, representative features are extracted from the fault evolution sample set, including key parameters such as current fluctuation patterns, temperature rise rate, displacement response delay, and electromagnetic noise disturbance modes. Secondly, feature points of corresponding dimensions are identified in the time series composed of trend information, and each feature is matched with a sample feature to establish a mapping relationship between time index and feature value. Thirdly, in the process of establishing the correspondence, each feature point in the trend series is matched with an indicator of the same dimension in the sample feature library, and the matching results are stored in the form of a matrix, so that the row dimension corresponds to the time index and the column dimension corresponds to the feature category. Finally, the resulting comparison matrix not only maintains the one-to-one correspondence between trend information and sample features, but also intuitively shows the differences between the two in a two-dimensional structure, providing a structured comparison basis for subsequent pattern comparison and fault identification.

[0144] It should be noted that this matrix processing unifies the originally independent time series features and sample database features into the same data framework, thereby enabling quantifiable and operable comparison conditions.

[0145] Step S404: In the comparison matrix, the trend information and the fault evolution samples are compared segment by segment based on the time evolution order to select the candidate samples with the highest matching degree.

[0146] In this embodiment, the time-series segmented matching mechanism based on the alignment matrix is ​​as follows: First, the alignment matrix is ​​divided into several continuous intervals according to the time index, and each segment contains the corresponding features of trend information and fault evolution samples. Second, the similarity between the trend information features and sample features is calculated in each segment to determine the degree of matching in that time period. Third, the matching results of each segment are accumulated in chronological order to form a segment-by-segment evolution matching curve, reflecting the fit between the trend information and different samples in the global time range. Finally, by comparing the cumulative matching curves of all candidate samples, the sample that maintains the highest consistency in the overall time dimension is selected and taken as the optimal candidate.

[0147] It should be noted that this segmented comparison method accumulates the matching degree gradually over time, avoiding the problem that a one-time overall comparison may ignore local differences. This allows the selected candidate samples to more accurately reflect the correspondence between trend information and actual fault modes, providing a reliable basis for the final fault type identification.

[0148] Step S405: Determine the fault type and severity of the electric actuator based on the annotation information of the candidate samples.

[0149] In this embodiment, based on the joint analysis of candidate sample annotation information and comparison results, specifically, firstly, the annotation information attached to the selected candidate samples is read, which typically includes the classification label of the fault type and the corresponding severity level; secondly, the trend information is cross-validated with the matching results of the candidate samples in the comparison matrix to confirm whether the fault mode annotated by the sample is highly consistent with the current trend characteristics; thirdly, according to the severity range defined in the annotation, the offset of the trend information on key features is correlated with the level standard to determine whether the fault is in a minor, progressive, or severe stage; finally, the judgment results of fault type and severity are combined to form a final diagnostic conclusion, which is then transmitted as output to the subsequent decision-making stage.

[0150] It should be noted that this step, through the dual constraints of sample labeling and comparative verification, ensures that the diagnostic results not only come from the numerical results of pattern matching, but also have prior definitions from the sample library, thereby achieving a systematic judgment on the nature and risk level of electric actuator faults.

[0151] Example 2

[0152] Please see Figure 3 Another embodiment of the present invention provides: an electric actuator fault diagnosis system, comprising: a data acquisition module, a trajectory construction module, a fault information extraction module, and a discrimination module;

[0153] The data acquisition module is used to collect current, voltage, displacement, temperature rise and electromagnetic noise signals of the electric actuator during operation, forming a multi-dimensional operation dataset;

[0154] The trajectory construction module is used to obtain feature vectors by time-frequency domain fusion processing of the multi-dimensional operating data, and to construct the health status trajectory of the electric actuator.

[0155] The fault information extraction module is used to introduce an electromagnetic noise coupling factor into the health state trajectory, perform evolution analysis on the health state trajectory, and extract trend information related to the fault.

[0156] The discrimination module is used to determine the fault type and severity by comparing the trend information with pre-stored actuator fault evolution samples when the trend information exceeds a set threshold.

[0157] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0158] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of diagnosing a failure of an electric actuator, characterized by, The application relates to a health state trajectory construction method of an electric actuator, and belongs to the technical field of electric actuator health state monitoring. The method comprises the following steps: Collecting current, voltage, displacement, temperature rise and electromagnetic noise signals of an electric actuator during operation to form a multi-dimensional operation data set; Processing the multi-dimensional operation data through time-frequency domain fusion to obtain a feature vector, and constructing a health state trajectory of the electric actuator; Introducing an electromagnetic noise coupling factor into the health state trajectory to perform evolution analysis on the health state trajectory, and extracting trend information related to faults; When the trend information exceeds a set threshold, comparing the trend information with pre-stored actuator fault evolution samples to identify the fault type and severity; Processing the multi-dimensional operation data through time-frequency domain fusion to obtain a feature vector, and constructing a health state trajectory of the electric actuator, which comprises the following steps: Mapping the multi-dimensional operation data to time domain, frequency domain and time-frequency joint domain respectively to form three types of basic signal representations; Introducing an association weight between actuator driving current and electromagnetic noise during feature coding of the three types of basic signal representations; Fusing the feature coding results in a sequence splicing manner to obtain a composite feature matrix containing dynamic and static information; In the composite feature matrix, a state vector group is extracted based on rearrangement of a sliding time slice; 2. The electric actuator fault diagnostic method of claim 1, wherein, Arranging the state vector group in sequence to generate a health state trajectory of the electric actuator. The method for fusing the feature coding results in a sequence splicing manner to obtain a composite feature matrix containing dynamic and static information comprises the following steps: Orderly indexing each feature coding result according to the time domain, frequency domain and time-frequency joint domain of the source to form an initial feature sequence with domain labels; In the initial feature sequence, the initial feature sequence is rearranged according to an alternating mode of the time domain, frequency domain and time-frequency joint domain to obtain a multi-level interleaved arrangement sequence; Cyclically splicing the arrangement sequence in the time dimension; 3. A method of fault diagnosis of an electric actuator according to claim 2, c h a r a c t e r i z e d b y Reconstructing the cyclically spliced arrangement sequence into a two-dimensional matrix, and the two-dimensional matrix is a composite feature matrix containing dynamic and static features. In the composite feature matrix, a state vector group is extracted based on rearrangement of a sliding time slice, which comprises the following steps: Setting a preset window length and a step along the time dimension of the composite feature matrix to generate a sliding time slice sequence; In each time slice, a joint reference column containing current and electromagnetic noise coding is selected as an anchor reference, and based on the time sequence marker of the anchor reference, the bit sequence of the to-be-aligned feature column in the time slice is rearranged to form an aligned sub-matrix; According to the dynamic marker and the static marker, the aligned sub-matrix is hierarchically split, the de-interleaving processing is first performed in the column dimension, and then the convergence operation is performed in the row dimension to obtain a candidate vector group of the time slice; 4. The electric actuator fault diagnostic method of claim 3, wherein, The candidate vector group is connected in a preset order to generate a state vector corresponding to the time slice, and the state vectors are arranged in sequence according to the time slice index identification to form the state vector group. The method for arranging the state vector group in sequence to generate a health state trajectory of the electric actuator comprises the following steps: According to the index identification of the state vectors in the state vector group in different time slices, the state vectors are time-sequentially aligned; The aligned state vectors are sequentially connected in a head-to-tail manner to form a continuous vector sequence. pathing mapping the continuous vector sequence in a multi-dimensional coordinate system; arranging the continuous vector sequence as a single trajectory according to the pathing mapping result, and taking the single trajectory as a health state trajectory of the electric actuator.

5. The electric actuator fault diagnostic method of claim 1, wherein, introducing an electromagnetic noise coupling factor into the health state trajectory, performing evolution analysis on the health state trajectory, and extracting trend information related to a fault, including: jointly calibrating the electromagnetic noise and the current based on the running signal of the electric actuator to generate the electromagnetic noise coupling factor; embedding the electromagnetic noise coupling factor into the health state trajectory as an additional dimension to form an extended trajectory with a coupling feature; fragmenting the extended trajectory according to a preset time window to generate trajectory fragments; comparing the trajectory fragments according to the change mode of the electromagnetic noise coupling factor to screen out abnormal offset fragments; arranging and merging the abnormal offset fragments in chronological order to extract trend information related to a potential fault.

6. A method of fault diagnosis of an electric actuator according to claim 5, c h a r a c t e r i z e d b y comparing the trajectory fragments according to the change mode of the electromagnetic noise coupling factor to screen out abnormal offset fragments, including: extracting the change mode of the electromagnetic noise coupling factor from each trajectory fragment to form a corresponding mode sequence; pairing the mode sequence according to the time index one by one, and performing contrastive analysis on the change modes between different trajectory fragments; identifying fragment positions with significant differences in the change modes in the contrastive analysis to form a candidate abnormal fragment set; performing secondary comparison on the candidate abnormal fragment set to screen out trajectory fragments with an offset degree exceeding a set threshold, and taking the trajectory fragments as abnormal offset fragments.

7. A method of fault diagnosis of an electric actuator according to claim 6, c h a r a c t e r i z e d b y arranging and merging the abnormal offset fragments in chronological order to extract trend information related to a potential fault, including: arranging the abnormal offset fragments in chronological order according to the time index to form an initial sequence; continuously splicing adjacent time fragments in the initial sequence to eliminate time gaps and form a local extended sequence; performing cross-segment fusion on the local extended sequence to integrate abnormal fragments distributed in different time periods to generate a global sequence; unfolding the global sequence according to a time axis to reconstruct a single time sequence, and taking the single time sequence as trend information related to a potential fault.

8. The electric actuator fault diagnostic method of claim 1, wherein, when the trend information exceeds a set threshold, discriminating a fault type and a severity by comparing with a pre-stored actuator fault evolution sample, including: detecting the value of the trend information, and starting a comparison process when the value exceeds a preset threshold; calling an actuator fault evolution sample set corresponding to the current running condition from a pre-stored database; establishing a one-to-one mapping relationship between the trend information and key features of the fault evolution sample to form a comparison matrix; performing a segment-by-segment comparison between the trend information and the fault evolution sample in the comparison matrix based on a time evolution order to screen out a candidate sample with the highest matching degree; determining the fault type and the severity of the electric actuator according to the labeling information of the candidate sample.

9. A system for diagnosing faults in an electric actuator, for implementing a method for diagnosing faults in an electric actuator according to any one of claims 1-8, characterized by including: a data acquisition module, a trajectory construction module, a fault information extraction module, and a discrimination module; The data acquisition module is configured to acquire current, voltage, displacement, temperature rise and electromagnetic noise signals of the electric actuator during operation, and form a multi-dimensional operation data set; The trajectory construction module is configured to obtain a feature vector by performing time-frequency domain fusion processing on the multi-dimensional operation data, and construct a health state trajectory of the electric actuator; The fault information extraction module is configured to introduce an electromagnetic noise coupling factor into the health state trajectory, perform evolution analysis on the health state trajectory, and extract trend information related to a fault; The discrimination module is configured to, when the trend information exceeds a set threshold, compare the trend information with pre-stored actuator fault evolution samples, and discriminate a fault type and a severity.

Images