Method and system for fault diagnosis of a crane motor

Abstract

The present application relates to the technical field of crane fault detection, and more particularly to a fault diagnosis method and system for a crane motor. The method comprises: acquiring motor operation data in real time through a sensor, extracting time domain and frequency domain statistical features, and generating a working condition classification label; determining an operation mode according to the working condition classification label, determining a deviation vector of a standard health sequence and an abnormal deviation threshold; calculating a similarity matrix of a current and historical sequence, optimizing a weight set; forming a comprehensive diagnosis index, and determining whether a potential fault signal is generated; if a fault signal exists, updating a diagnosis model parameter, forming an optimized diagnosis framework, and analyzing operation data based on the framework to obtain a final fault diagnosis result. The present application solves the problems of low fault recognition accuracy, diagnosis lag and non-self-adaptive model of the crane motor under multiple working conditions, and realizes intelligent monitoring and dynamic precise diagnosis of the motor operation state.

Description

Technical Field

[0001] This invention relates to the field of crane fault detection technology, and in particular to a fault diagnosis method and system for crane motors. Background Technology

[0002] As a key piece of equipment in industrial production and logistics transportation, the operating status of the core drive component of cranes—the motor—directly affects the safety, reliability, and production efficiency of the equipment. During long-term operation, crane motors are often subjected to high loads, frequent starts and stops, and complex electromagnetic environments, making them highly susceptible to various faults such as winding overheating, rotor imbalance, bearing wear, and stator core loosening. If these faults are not detected and identified in a timely manner, they can not only lead to a decline in equipment performance but may even cause serious safety accidents such as crane loss of control and heavy-load falls, resulting in economic losses and personal injury. Therefore, how to perform real-time and accurate fault diagnosis of crane motors is an important technical issue in the field of motor health monitoring and intelligent maintenance.

[0003] Traditional motor fault diagnosis methods often rely on manual inspection or analysis based on single signal features, such as spectral analysis of current signals or envelope analysis of vibration signals. These methods suffer from strong reliance on experience, long diagnostic delays, limited feature information, and poor anti-interference capabilities. Especially in the complex operating environment of cranes, signals are easily affected by electromagnetic interference, load fluctuations, and mechanical shocks, leading to unstable diagnostic results. Furthermore, traditional models typically rely on static threshold judgments, making it difficult to handle the dynamic changes in the characteristic distribution of motors under different operating conditions (such as no-load, rated load, and overload), resulting in low accuracy in multi-condition identification.

[0004] To address the aforementioned shortcomings, this invention proposes a fault diagnosis method for crane motors. This method constructs a complete diagnostic process encompassing data acquisition, feature extraction, operating condition identification, deviation modeling, similarity analysis, signal fusion, and model self-optimization. By combining real-time sensor data with historical operating samples, it dynamically adjusts model parameters and anomaly thresholds to achieve accurate identification and adaptive judgment of motor faults under different operating conditions. This method not only improves the real-time performance and accuracy of fault diagnosis but also enhances the system's robustness and scalability under complex load environments, making it suitable for intelligent operation and maintenance scenarios for cranes and other heavy-duty motor equipment. Summary of the Invention

[0005] This invention provides a fault diagnosis method and system for crane motors. By constructing a multi-level diagnostic process that includes data acquisition, feature extraction, deviation identification, similarity analysis, signal fusion, and model optimization, it enables intelligent and accurate fault identification of crane motors under complex working conditions.

[0006] In a first aspect, the present invention provides a fault diagnosis method for a crane motor, comprising:

[0007] Step S1: Acquire crane motor operation data in real time through sensors to obtain an operation sequence; obtain operating condition classification labels based on the time-domain statistical features and frequency-domain statistical features extracted from the operation sequence;

[0008] Step S2: Determine the operating mode of the crane motor based on the operating condition classification label, determine the deviation vector between the operating sequence and the standard healthy sequence through the support vector machine model, and calculate the abnormal deviation threshold; obtain the similarity matrix between the current operating sequence and the historical sequence based on the abnormal deviation threshold, adjust the similarity matrix through the dynamic programming algorithm, and obtain the weight set of the adjusted similarity matrix;

[0009] Step S3: By integrating the running data through the adjusted weight set, a comprehensive diagnostic index is formed. Based on the comprehensive diagnostic index, the comprehensive diagnostic index is obtained, and it is determined whether the comprehensive diagnostic index exceeds the threshold to obtain potential fault signals.

[0010] Step S4: Update the diagnostic model parameters based on the potential fault signals to obtain an optimized diagnostic framework; analyze the operating data based on the optimized diagnostic framework, fuse the operating condition classification label and the deviation vector to obtain the final fault diagnosis result.

[0011] As a preferred embodiment of the present invention, step S1, obtaining the running sequence, includes:

[0012] The current, voltage, and speed signals of the motor are acquired by sensors; the current, voltage, and speed signals are processed to filter out noise and obtain an operating sequence; the operating sequence is analyzed in the time domain to extract time series features; and the operating sequence is analyzed in the frequency domain to extract frequency distribution features.

[0013] As a preferred embodiment of the present invention, in step S1, obtaining the working condition classification label includes:

[0014] The time-domain statistical features and frequency-domain statistical features of the running sequence are combined into a feature vector; the feature vector is grouped and analyzed by a pre-trained clustering model to obtain an initial classification result, wherein the time-domain statistical features include mean, variance and peak value, and the frequency-domain statistical features include power spectral density and dominant frequency component;

[0015] The initial classification result is matched with a preset operating condition template to generate the operating condition classification label, wherein the operating condition classification label represents the operating status of the motor.

[0016] As a preferred embodiment of the present invention, step S2, obtaining the abnormal deviation threshold, includes:

[0017] If the operating condition classification label indicates a high-load mode, then the pre-trained support vector machine model is activated; the running sequence is analyzed by the support vector machine model to generate a deviation vector, wherein the deviation vector represents the degree of deviation of the running sequence;

[0018] Based on the distribution characteristics of the deviation vector, an abnormal deviation threshold is calculated; the abnormal deviation threshold is verified, and the classification boundary is adjusted; the abnormal deviation threshold is updated based on the adjusted classification boundary, wherein the abnormal deviation threshold is used to determine operational abnormalities.

[0019] As a preferred embodiment of the present invention, step S2, obtaining the adjusted weight set, includes:

[0020] Obtain a historical running sequence dataset; based on the abnormal deviation threshold, filter out reference sequences from the historical running sequence dataset; calculate the similarity between the current running sequence and the reference sequences, and generate a similarity matrix;

[0021] The weight coefficients of the similarity matrix are optimized using a dynamic programming algorithm to obtain an initial weight set; the initial weight set is then normalized to generate an adjusted weight set.

[0022] As a preferred embodiment of the present invention, step S3, obtaining the potential fault signal, includes:

[0023] Acquire the current signal and speed signal from the running sequence; perform weighted fusion of the current signal and speed signal according to the adjusted weight set to generate a fused signal; perform feature extraction on the fused signal to obtain a comprehensive feature vector;

[0024] A comprehensive diagnostic index is calculated based on the comprehensive feature vector, wherein the comprehensive diagnostic index represents the operating health status of the motor; it is determined whether the comprehensive diagnostic index exceeds a preset diagnostic threshold, and a potential fault signal is generated based on the determination result.

[0025] As a preferred embodiment of the present invention, step S4, obtaining the optimized diagnostic framework, includes:

[0026] If the potential fault signal exists, real-time load feature data is acquired; the parameters of the pre-trained diagnostic model are updated based on the real-time load feature data; the threshold boundary of the diagnostic model is adjusted through an iterative optimization algorithm to generate an optimized threshold boundary; the classification rules of the diagnostic model are updated based on the optimized threshold boundary; and an optimized diagnostic framework is generated based on the updated classification rules.

[0027] As a preferred embodiment of the present invention, step S4, obtaining the final fault diagnosis result, includes:

[0028] The process involves: acquiring subsequent operational data; extracting features from the subsequent operational data using the optimized diagnostic framework to generate subsequent feature vectors; matching the subsequent feature vectors with the operating condition classification labels to generate classification results; fusing the classification results with the deviation vector to generate a comprehensive diagnostic vector; and comparing the comprehensive diagnostic vector with a preset fault template to obtain the final fault diagnosis result, wherein the final fault diagnosis result characterizes the fault type of the motor.

[0029] Secondly, the present invention also provides a fault diagnosis system for a crane motor, for implementing the above-described method, the system comprising:

[0030] The operating condition classification unit is used to acquire crane motor operating data in real time through sensors to obtain an operating sequence; and to obtain an operating condition classification label based on the time-domain statistical features and frequency-domain statistical features extracted from the operating sequence.

[0031] An abnormal deviation calculation unit is used to determine the operating mode of the crane motor based on the operating condition classification label, determine the deviation vector between the operating sequence and the standard healthy sequence through a support vector machine model, and calculate the abnormal deviation threshold.

[0032] The similarity analysis unit is used to obtain the similarity matrix between the current running sequence and the historical sequence based on the abnormal deviation threshold, adjust the similarity matrix through a dynamic programming algorithm, and obtain the weight set of the adjusted similarity matrix.

[0033] The fault diagnosis unit is used to fuse operational data through an adjusted weight set to form a comprehensive diagnostic index. Based on the comprehensive diagnostic index, it obtains a comprehensive diagnostic index, determines whether the comprehensive diagnostic index exceeds a threshold, and obtains potential fault signals. It updates the diagnostic model parameters according to the potential fault signals to obtain an optimized diagnostic framework. Based on the optimized diagnostic framework, it analyzes operational data, fuses the operating condition classification label and the deviation vector, and obtains the final fault diagnosis result.

[0034] Thirdly, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0035] The beneficial effects of this invention are as follows:

[0036] This invention acquires current, voltage, and speed signals from a motor in real time using sensors, generating an operating sequence. After filtering, normalization, and time-frequency domain feature extraction, a high-quality data foundation is provided for subsequent analysis. This sequence reflects both the dynamic response of the motor and retains load fluctuation characteristics, providing rich information support for operating condition classification. A clustering model is used to group and analyze the feature vectors of the operating sequence, and operating condition classification labels are generated using preset operating condition templates, enabling automatic identification of the motor's operating mode and providing accurate status indicators for subsequent deviation calculations. Based on the operating condition identification results, a support vector machine model is used to calculate the deviation vector of the operating sequence relative to a standard healthy sequence, and an abnormal deviation threshold is determined accordingly. Through threshold verification and boundary optimization, it can adapt to different noise environments and load conditions, thereby obtaining a more stable anomaly detection capability. Based on this threshold, a reference sequence is selected from historical operating sequences, and dynamic programming is used to calculate... The method calculates the similarity between the current sequence and the reference sequence, optimizes the weight set, and achieves the fusion of multi-dimensional data. The fused signal is further extracted to extract comprehensive features, and a comprehensive diagnostic index is calculated through the model to determine whether there are potential fault signals. When a potential anomaly is detected, the model parameters are updated using real-time load feature data, and the classification boundary is adjusted through an iterative optimization algorithm to generate an optimized diagnostic framework. The optimized framework performs feature analysis and classification matching on newly collected operating data, combines the deviation vector to form a comprehensive diagnostic vector, and compares it with a preset fault template to output the final fault type. Through the cooperation of the above technical solutions, dynamic diagnosis and adaptive learning of crane motors under multiple working conditions and multiple signal environments are realized. It can not only accurately identify early anomalies, but also optimize the model in real time according to the operating status, realizing a closed-loop diagnosis from data acquisition to intelligent decision-making, which significantly improves the accuracy, real-time performance, and robustness of fault identification. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of a fault diagnosis method for a crane motor in an embodiment.

[0039] Figure 2 This is a structural diagram of a fault diagnosis system for a crane motor in an embodiment;

[0040] Figure 3 The diagram shows the relationship between the load and speed deviation of the wind turbine generator in the embodiment. Detailed Implementation

[0041] This invention provides a method and system for fault diagnosis of crane motors. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0042] For ease of understanding, the specific process of the embodiments of the present invention will be described below, such as... Figure 1 As shown, the fault diagnosis method for crane motors in this embodiment of the invention includes:

[0043] Step S1: Acquire crane motor operation data in real time through sensors to obtain an operation sequence; obtain operating condition classification labels based on the time-domain statistical features and frequency-domain statistical features extracted from the operation sequence;

[0044] In step S1, obtaining the running sequence includes: acquiring the motor's current signal, voltage signal, and speed signal through sensors; performing signal processing on the current signal, voltage signal, and speed signal to filter out noise and obtain the running sequence; performing time-domain analysis on the running sequence to extract time series features; and performing frequency-domain analysis on the running sequence to extract frequency distribution features.

[0045] Specifically, during the operation of the crane motor, current, voltage, and speed signals are collected in real time by current sensors, voltage sensors, and speed sensors installed on the motor body. Each sensor is electrically connected to the data acquisition module and input to the data acquisition unit via a data synchronization interface. The multi-source signals are synchronously sampled at a preset sampling frequency to ensure that the dynamic operation of the motor under different loads and start-stop states can be completely captured, forming a raw signal dataset. To ensure the effectiveness and stability of the acquired signals, the data acquisition unit first performs a signal preprocessing process after the signal input. The raw current, voltage, and speed signals are denoised by digital filtering, i.e., low-pass filtering is used to filter out high-frequency interference components. By adjusting the cutoff frequency and order parameters of the filter, the signal characteristics of the crane motor under high load fluctuations or frequent start-stop conditions are adapted to obtain smoothed current, voltage, and speed data. These data are then combined into a first operating sequence with time as the sequence axis. The above operating sequence continuously reflects the relationship between the motor's operating state and time, providing basic data for subsequent feature extraction.

[0046] Based on this, time-domain analysis is performed on the above operating sequence. Statistical feature parameters of the sequence are calculated through feature extraction, including mean, variance, peak value, and their changing trends, to characterize the stability and load fluctuation characteristics during motor operation. Specifically, the time-domain feature extraction module divides the operating sequence according to sampling windows, calculates features for the data within each window, and updates them in a sliding manner to obtain a time-domain statistical feature set covering the entire operating cycle. Frequency-domain analysis is also performed on the above operating sequence, that is, the time-series signal is converted from the time domain to the frequency domain using a fast Fourier transform algorithm to extract spectral feature parameters. Through spectral amplitude and frequency distribution analysis, the dominant frequency component and energy distribution characteristics during motor operation can be identified, thereby revealing potential mechanical vibration characteristics, electromagnetic imbalance phenomena, or abnormal load modes. The extraction of frequency-domain features complements the time-domain features, jointly reflecting the comprehensive operating state of the motor. The combination of time-domain and frequency-domain features not only preserves the time-varying trend information but also reflects the spectral distribution characteristics, making the multi-dimensional representation of the motor's operating state more complete.

[0047] The above operating sequence logically establishes a mapping relationship between multi-source sensor data and motor operating state characteristics. The current signal corresponds to changes in electrical load, the voltage signal reflects electromagnetic drive characteristics, and the speed signal reveals the mechanical motion state. After fusion, the three form an operating data structure with temporal continuity and spectral integrity, providing accurate input for subsequent working condition identification, deviation calculation, and model optimization, thereby significantly improving the accuracy and robustness of crane motor fault diagnosis.

[0048] Further, in step S1, the operating condition classification labels are obtained, including:

[0049] The time-domain statistical features and frequency-domain statistical features of the running sequence are combined into a feature vector; the feature vector is grouped and analyzed by a pre-trained clustering model to obtain an initial classification result, wherein the time-domain statistical features include mean, variance and peak value, and the frequency-domain statistical features include power spectral density and dominant frequency component;

[0050] The initial classification result is matched with a preset operating condition template to generate the operating condition classification label, wherein the operating condition classification label represents the operating status of the motor.

[0051] Specifically, by jointly analyzing and pattern recognizing the time-domain and frequency-domain features of the operating sequence, intelligent classification of the motor's operating state is achieved. First, based on the obtained standardized operating sequence, statistical analysis is performed on current, voltage, and speed signals to obtain time-domain feature parameters such as mean, variance, and peak value. The mean characterizes the overall average level of motor operation, the variance reflects load fluctuations or operational stability, and the peak value reveals potential instantaneous impacts or overloads during operation. Statistical calculations of these signals over time capture the energy variation patterns of the crane motor under different operating conditions, providing a foundation for subsequent frequency-domain analysis. Furthermore, by converting the time-domain signal from the operating sequence to the frequency domain using a Fast Fourier Transform algorithm, frequency-domain features such as power spectral density and dominant frequency components are extracted. Power spectral density describes the signal energy at different frequencies. The distribution of features across dimensions reveals the characteristics of changes in the internal magnetic field and load cycle fluctuations of the motor. The dominant frequency component represents the main excitation frequency during motor operation, reflecting the imbalance or mechanical resonance of rotating parts. Through the mutual complementation of time-domain and frequency-domain features, the former reflects the overall trend and stability of motor operation, while the latter reveals the internal energy structure and vibration characteristics, thus jointly constituting a complete characterization of the motor's operating state. After completing the above feature extraction, the above time-domain statistical features and frequency-domain statistical features are fused to generate a unified feature vector. The feature vector is expressed in the form of a multi-dimensional array, with each dimension corresponding to different statistical feature parameters. The time information and frequency information are synergistically expressed through feature concatenation. To avoid the influence of deviations in the dimensions or numerical ranges of different features on the clustering results, normalization is performed before feature fusion to make each feature dimension comparable at the same scale, thereby ensuring the stability and reliability of subsequent algorithm processing.

[0052] Based on the generated feature vectors, a pre-trained clustering model is invoked to perform group analysis on the feature vector set. This clustering model can be an unsupervised learning model based on the K-means algorithm, using Euclidean distance to measure the similarity between feature vectors. Through an iterative optimization process, multiple cluster centers are determined, and feature vectors with high similarity are grouped into the same cluster. This clustering process is essentially an automatic partitioning of the feature space of the operating sequence, enabling operating data with similar statistical characteristics to be identified as having the same operating mode. During the training phase, the clustering model uses historical motor operating data as a sample set. By learning from operating samples under different loads, start-stop frequencies, and temperature conditions, it obtains cluster centers that can distinguish between normal, light load, rated, high load, and abnormal operating conditions. Therefore, during the diagnostic phase, it can be directly used for operating condition identification of the current operating data. The historical motor operation data includes the time-domain and frequency-domain characteristics of the motor's current, voltage, and speed data. After completing the clustering analysis, the initial classification results are matched with preset operating condition templates to generate the final operating condition classification labels. The aforementioned operating condition templates are standard models constructed based on expert experience and historical diagnostic results, used to characterize the feature distribution of various typical operating states. The templates contain reference intervals and threshold definitions for time-domain and frequency-domain features. For example, high-load templates typically correspond to higher peak values, larger variances, and spectral features with concentrated high-frequency energy. By calculating the similarity between the initial clustering results and each template, for example, using the cosine similarity method to measure the angle between feature vectors, the degree of matching between the current operating state and the template is determined. When the similarity exceeds the preset threshold, the current operating sequence is determined to correspond to the operating condition type of that template.

[0053] The successfully matched templates are output as operating condition classification labels to characterize the current operating status of the crane motor. These operating condition classification labels not only reflect whether the motor is in normal, low-load, or high-load mode, but also identify potential operating conditions with abnormal fluctuations. The above technical solution, through a multi-level processing flow of time-frequency feature extraction, feature fusion, cluster analysis, and template matching, logically realizes a complete path from data feature extraction to operating condition label generation. This transforms the crane motor operating status identification process from experience-based judgment to data-driven algorithm decision-making, significantly improving the accuracy and robustness of operating condition classification and providing a reliable input basis for subsequent deviation analysis and fault diagnosis.

[0054] Step S2: Determine the operating mode of the crane motor based on the operating condition classification label, determine the deviation vector between the operating sequence and the standard healthy sequence through the support vector machine model, and calculate the abnormal deviation threshold; obtain the similarity matrix between the current operating sequence and the historical sequence based on the abnormal deviation threshold, adjust the similarity matrix through the dynamic programming algorithm, and obtain the weight set of the adjusted similarity matrix;

[0055] In step S2, the abnormal deviation threshold is obtained, including:

[0056] If the operating condition classification label indicates a high-load mode, then the pre-trained support vector machine model is activated; the running sequence is analyzed by the support vector machine model to generate a deviation vector, wherein the deviation vector represents the degree of deviation of the running sequence;

[0057] Based on the distribution characteristics of the deviation vector, an abnormal deviation threshold is calculated; the abnormal deviation threshold is verified, and the classification boundary is adjusted; the abnormal deviation threshold is updated based on the adjusted classification boundary, wherein the abnormal deviation threshold is used to determine operational abnormalities.

[0058] Specifically, to achieve abnormal offset identification and dynamic threshold determination during motor operation, thereby improving the accuracy and robustness of subsequent diagnostic models under high load and complex operating conditions, when the system indicates that the motor is in a high load mode based on the operating condition classification label obtained from the above feature analysis, the control module automatically activates a pre-trained support vector machine model to perform targeted analysis on the current operating sequence. This support vector machine model is trained using a historical high-load operating dataset. By learning a large number of signal distribution features under normal and abnormal operating conditions, a classification hyperplane capable of distinguishing the degree of operating offset is established to identify the offset of the current operating state relative to the standard healthy sequence. The aforementioned historical high-load operating dataset includes historical high-load state operating sequences, standard healthy sequences, and corresponding deviation vectors. During execution, the filtered... Current, speed, and voltage signals are used as inputs to the aforementioned support vector machine (SVM) model. The SVM model first maps the input data to a high-dimensional feature space to capture nonlinear relationship features. It then determines the optimal separating hyperplane in this high-dimensional space using a kernel function, such as a radial basis function kernel. Subsequently, the SVM model calculates the deviation between the input sequence and the standard healthy sequence in the training set and outputs a multi-dimensional deviation vector in numerical form. Each dimension corresponds to a different physical quantity, such as the offset amplitude of the current channel or the fluctuation degree of the speed channel, thereby achieving a quantitative representation of the motor's multi-signal offset characteristics. This deviation vector not only reflects the distance relationship between the motor's operating sequence and the standard state in the overall feature space but also reveals the offset coupling characteristics between different physical signals, providing data support for subsequent threshold determination.

[0059] After obtaining the deviation vector, its statistical distribution characteristics are further analyzed. By calculating the sample mean and standard deviation of each dimension of the deviation vector, an abnormal deviation threshold is calculated based on the distribution characteristics. For example, the abnormal deviation threshold is equal to the mean plus three times the standard deviation, used to characterize the boundary range between normal and abnormal deviations. To ensure the stability of the above abnormal deviation threshold under different noise environments and load fluctuation conditions, the generated deviation threshold is verified and dynamically optimized. Specifically, the abnormal deviation threshold is applied to the validation dataset to calculate the accuracy. If the accuracy is lower than the preset value, the classification boundary of the support vector machine model is adjusted. The support vector positions are optimized, and the verification is iteratively performed until the accuracy reaches the target. For example, in high-load motor diagnostics, the above adjustments can be made for different noise levels. In noisy factory environments, boundary adjustments can reduce false positives and improve threshold robustness. In continuously operating pump motor scenarios, the verification process can use historical fault data, and the threshold is more accurate after boundary adjustments, resulting in early fault warnings. It should be noted that adjusting the classification boundary can dynamically adapt to changes in operating conditions and ensure the effectiveness of the threshold. For example, in elevator motor applications, if the deviation vector distribution is wide, the verification can refine the boundary and optimize the threshold to cover sudden load peaks. The abnormal deviation threshold is updated according to the adjusted classification boundary, wherein the abnormal deviation threshold is used to determine operational abnormalities.

[0060] Through the aforementioned iterative optimization process, the abnormal deviation threshold can be dynamically and adaptively adjusted according to different operating conditions and noise levels, maintaining high stability and sensitivity even in crane motor application scenarios with high load, strong vibration, and large signal fluctuations. Finally, the deviation threshold is recalculated based on the optimized classification boundary and updated in the diagnostic module for real-time judgment of operational anomalies. When the deviation vector generated after the new input operating data is analyzed by the model exceeds the aforementioned deviation threshold, it is determined that there is a potential abnormal state in the motor operation, and the subsequent fault signal analysis module is triggered. The above technical solution, by introducing a deviation modeling and threshold adaptive optimization mechanism based on support vector machines, logically realizes a continuous data flow processing process from operating condition identification to anomaly judgment. It can effectively identify small abnormal deviations of crane motors under high load and complex environments, thereby providing early warning of potential faults and improving the reliability and practicality of the overall fault diagnosis system.

[0061] Further, in step S2, the adjusted weight set is obtained, including:

[0062] Obtain a historical running sequence dataset; based on the abnormal deviation threshold, filter out reference sequences from the historical running sequence dataset; calculate the similarity between the current running sequence and the reference sequences, and generate a similarity matrix;

[0063] The weight coefficients of the similarity matrix are optimized using a dynamic programming algorithm to obtain an initial weight set; the initial weight set is then normalized to generate the adjusted weight set.

[0064] Specifically, in order to achieve weighted optimization of multi-source signals through historical data screening and similarity analysis, thereby improving the accuracy and dynamic adaptability of motor operating status determination, a historical operating sequence dataset is first obtained from the database. The dataset consists of long-term monitored current signals, voltage signals, and speed signals. Each sequence corresponds to the operating status of the motor under different operating conditions. After preprocessing and time synchronization, the historical sequences are formed into a standardized signal set that can be directly used in calculations, providing a data foundation for subsequent reference screening and similarity calculation.

[0065] Subsequently, based on the abnormal deviation threshold calculated using the aforementioned support vector machine model, the historical operating sequence dataset is filtered. This abnormal deviation threshold reflects the boundary between normal and abnormal operating states, enabling the identification of the stable boundary of motor operation under high load conditions. By traversing each sequence in the historical dataset, the matching degree between its deviation vector and the abnormal deviation threshold is calculated. If the deviation amplitude of the sequence is lower than the threshold, it indicates that the sequence belongs to a healthy or normal high-load operating condition, and it is selected as a reference sequence. Through this filtering method, invalid data containing noise or abnormal fluctuations are eliminated, and a subset of sequences that can represent typical operating states are retained, thus forming a reference sequence set. This reference sequence set statistically reflects the standard response characteristics of the motor under specific operating modes, ensuring... Subsequent similarity calculations demonstrate high reliability and representativeness. After obtaining the reference sequences, the similarity between the current running sequence and each reference sequence is calculated to quantify the proximity of the current running state to historical health states. Specifically, the multidimensional signals are matched using the Euclidean distance formula or other similarity measurement algorithms. Each reference sequence is compared item by item with the current running sequence to obtain a similarity value. All calculation results are then output in matrix form, forming a similarity matrix. Each element of the similarity matrix represents the degree of matching between the corresponding reference sequence and the current sequence; a higher value indicates that the running state is closer to the standard health mode. Through this similarity matrix, a global matching relationship between time series is realized in the data structure, providing a quantifiable basis for subsequent weight allocation. For example, as shown... Figure 3 As shown, in the wind turbine scenario, the deviation calculation of the speed sequence shows that when the load increases, the deviation vector value increases from 0.2 to 1.5. The elements in the deviation vector are the ratio of the error value to the standard value, and the deviation value is the modulus of the deviation vector. The larger the deviation value, the greater the possibility of failure, indicating potential bearing wear. This method improves the accuracy of diagnosis.

[0066] Based on the aforementioned similarity matrix, a dynamic programming algorithm is applied to optimize the matrix weight coefficients to determine the contribution of different reference sequences to the comprehensive diagnostic results. The dynamic programming algorithm uses the similarity matrix as the state transition space, recursively calculating the optimal path for cumulative similarity to determine the optimal weight allocation scheme. Each row and column intersection in the similarity matrix is ​​considered a state node, and the optimal cumulative weight of the current node is calculated based on the similarity value and the cumulative weight of the previous state, thus avoiding local optima and obtaining the globally optimal path. Solving the dynamic programming algorithm yields an initial set of weights, where each weight value corresponds to the contribution of a specific reference sequence. High-similarity sequences receive larger weights, while sequences with significant differences from the current operating state have their weights automatically reduced. This dynamic programming optimization achieves intelligent weighting of multi-source data under complex motor operating conditions, making the weighting results more reflective of the actual operating characteristics of the motor. After completing the dynamic optimization, the initial weights are... The weight set is normalized to ensure that the sum of all weight values ​​is constant, thus maintaining the consistency of the fusion calculation ratio. Normalization is achieved by dividing each weight value by the sum of the weights in the set, allowing the weights of each reference sequence to directly participate in subsequent signal fusion calculations on a relative scale. The adjusted weight set generated after normalization possesses mathematical regularity and physical interpretability, remains numerically stable, and dynamically reflects the impact of different signals on diagnostic results under changing operating conditions. This technical solution achieves correlation and quantization mapping between different times, loads, and signal sources at the data level, thereby constructing a highly robust and sensitive weight optimization mechanism. The final adjusted weight set can be used in subsequent multi-signal fusion steps, enabling the comprehensive diagnostic index to reflect current operating conditions while maintaining dynamic adaptability to historical health states, thus effectively improving the accuracy and stability of the crane motor fault diagnosis system in complex environments.

[0067] Step S3: By fusing the operational data through the adjusted weight set, a comprehensive diagnostic index is formed. Based on the comprehensive diagnostic index, the comprehensive diagnostic index is obtained, and it is determined whether the comprehensive diagnostic index exceeds the threshold to obtain potential fault signals; specifically including:

[0068] Acquire the current signal and speed signal in the running sequence; perform weighted fusion of the current signal and speed signal according to the adjusted weight set to generate a fused signal; perform feature extraction on the fused signal to obtain a comprehensive feature vector;

[0069] A comprehensive diagnostic index is calculated based on the comprehensive feature vector, wherein the comprehensive diagnostic index represents the operating health status of the motor; it is determined whether the comprehensive diagnostic index exceeds a preset diagnostic threshold, and a potential fault signal is generated based on the determination result.

[0070] Specifically, in order to achieve weighted fusion of multi-source operating signals and quantitative assessment of health status, thereby establishing a comprehensive diagnostic mechanism that can reflect the dynamic characteristics of motor operation, since voltage signal changes often cannot directly reflect the internal fault state of the motor, especially for mechanical and electromagnetic coupling faults such as bearing wear, rotor imbalance, winding short circuit or overload, the early symptoms are mainly reflected in current distortion and speed fluctuation, while voltage changes are not significant. Therefore, current and speed signals are first extracted from the aforementioned standardized operating sequence. These two types of signals are collected in real time by current and speed sensors installed on the motor body. After filtering and noise reduction processing, they are input to the fusion module. Among them, the current signal mainly reflects the electrical load and magnetic field change characteristics of the motor, while the speed signal characterizes the smoothness and inertial response of mechanical motion. Together, they constitute a dual representation of the motor's operating behavior.

[0071] After signal acquisition, the adjusted weight set obtained through dynamic programming in the previous stage is called to perform weighted fusion of the current and speed signals. Each weight value in the weight set represents the degree of influence of different signals on the overall operating state under the current operating conditions. For example, in high-load mode, the current signal has a higher weight to reflect load impact, while in variable speed operation mode, the speed signal has a larger weight to emphasize mechanical stability. During the fusion process, the weight values ​​corresponding to the current and speed signals in the weight set are first read, and each signal is weighted according to its corresponding weight. Then, the weighted signals are linearly combined to generate a fused signal. The fused signal retains the dynamic change characteristics of the original signal in the time domain, and reflects the comprehensive contribution of each signal to the overall state of the motor in the amplitude distribution, thus constructing an operating information expression with both electrical and mechanical characteristics. Through the above weighted fusion method based on weight optimization, the system realizes dynamic coordination of multi-source signals at a unified scale, avoids the one-sidedness caused by single signal analysis, and improves the stability and accuracy of subsequent diagnostic analysis.

[0072] Subsequently, feature extraction is performed on the generated fused signal to obtain a comprehensive feature vector that represents the motor's operating state. The feature extraction process adopts an analysis strategy combining the time domain and the frequency domain. In the time domain, statistical features such as the mean, variance, and peak value of the fused signal are calculated to characterize the signal's fluctuation trend and stability. In the frequency domain, the dominant frequency component and power distribution features are extracted through Fourier transform to reveal the frequency structure of the motor's operating energy and the periodicity of the load. The time-domain features and frequency-domain features are uniformly normalized and vector concatenated to form a complete comprehensive feature vector. The above feature vector establishes a mapping relationship between the fused signal and the motor's health state in terms of data structure. Each feature dimension corresponds to a different physical meaning, such as the mean reflecting the average power level, the variance reflecting the load fluctuation intensity, and the dominant frequency in the frequency domain reflecting the mechanical vibration characteristics, thereby achieving a comprehensive quantification of the operating health.

[0073] After obtaining the comprehensive feature vector, a comprehensive diagnostic index is calculated based on it. During the calculation, a pre-trained support vector machine (SVM) classifier is invoked to score the health status of the feature vector. The SVM classifier, based on the hyperplane separation principle in high-dimensional feature space, calculates the distance from the sample to the classification boundary through inner product operations, thereby generating a health status score. This score is then normalized to obtain the comprehensive diagnostic index. The value range is within a standardized interval; the closer the value is to the normal range, the more stable the operation; the closer it is to the upper limit, the more potential anomalies exist. The training data for the SVM classifier consists of historical comprehensive feature vectors and corresponding comprehensive diagnostic indices. Through this method, complex multidimensional signal features... The value is transformed into a single indicator, realizing the mapping from a multi-feature space to a health measurement space, enabling the motor operating status to be expressed in a quantitative form. The calculated comprehensive diagnostic indicator is further compared with a preset diagnostic threshold. If the indicator exceeds the diagnostic threshold, a potential fault signal is generated. The diagnostic threshold is a judgment boundary set based on historical operating data and an empirical model, used to distinguish between normal fluctuations and abnormal deviations. If the comprehensive diagnostic indicator exceeds the threshold, it means that the operating status represented by the current fused signal has deviated from the standard operating conditions, and there may be potential risks such as mechanical wear, abnormal current fluctuations, or load imbalance. When this situation is detected, a potential fault signal is immediately output to trigger the subsequent model adaptive optimization and maintenance early warning mechanism.

[0074] The aforementioned technical solution forms a closed-loop logical chain from signal fusion, feature extraction, index calculation to threshold judgment. Weighted fusion realizes dynamic coupling and optimized coordination among multi-source signals, feature extraction establishes a multi-dimensional mapping between the fused signal and the physical state, index calculation realizes a non-linear mapping from high-dimensional features to health metrics through support vector machines, and threshold judgment constitutes the decision-making link for diagnostic result output. Through this multi-layered data association and algorithm collaboration mechanism, the system can accurately diagnose the dynamic state of crane motors under complex operating conditions such as high load, variable speed, and long cycle operation, realizing intelligent processing of the entire process from signal to decision, and significantly improving the timeliness and accuracy of fault detection.

[0075] Step S4: Update the diagnostic model parameters based on the potential fault signals to obtain an optimized diagnostic framework; analyze the operating data based on the optimized diagnostic framework, fuse the operating condition classification label and the deviation vector to obtain the final fault diagnosis result;

[0076] In step S4, the optimized diagnostic framework is obtained, including:

[0077] If the potential fault signal exists, real-time load feature data is acquired; the parameters of the pre-trained diagnostic model are updated based on the real-time load feature data; the threshold boundary of the diagnostic model is adjusted through an iterative optimization algorithm to generate an optimized threshold boundary; the classification rules of the diagnostic model are updated based on the optimized threshold boundary; and the optimized diagnostic framework is generated based on the updated classification rules.

[0078] Specifically, in order to dynamically and adaptively adjust the pre-trained diagnostic model through a real-time data feedback mechanism, thereby forming an intelligent diagnostic framework that can continuously evolve and match changes in the current operating conditions, when the system detects a potential fault signal, it indicates that the motor's operating state has deviated from the standard healthy mode. At this time, the model update process is immediately initiated, that is, real-time load characteristic data related to the current operating conditions is first acquired. The aforementioned real-time load characteristic data is collected in real time by multiple sets of sensors deployed on the crane motor, including current signals, speed signals and their derived power load indicators, and after filtering and normalization processing, a real-time load characteristic dataset is formed, which not only reflects the dynamic behavior of the motor in the current operating cycle, but also directly corresponds to the potential fault signal in time.

[0079] After acquiring real-time load characteristic data, the parameters of the pre-trained diagnostic model are updated based on this data. The original diagnostic model is trained using a large amount of historical operating condition data to identify deviation patterns of the motor under different load conditions. During the model update phase, the real-time load characteristic data is input into the diagnostic model and compared with the feature distribution of the historical training data. The degree of deviation is calculated through difference analysis, and the diagnostic model parameters are dynamically adjusted accordingly. This adjustment process rebalances the model weights and optimizes the kernel function parameters, enabling the diagnostic model's discrimination boundary to adapt to the actual load changes of the motor in real time. For example, when a crane is lifting a heavy load, the peak current and speed fluctuations will increase. After adjusting the diagnostic model parameters, the sensitivity to high load anomalies will automatically increase, thereby reducing false positives and false negatives. Through this real-time parameter update mechanism, the diagnostic model can maintain classification accuracy and response sensitivity when facing nonlinear and non-stationary operating condition changes. The training data for the diagnostic model includes historical operating condition data and corresponding deviation patterns. The historical operating condition data includes historical current signals, speed signals, and power load indicators.

[0080] After the diagnostic model parameters are adjusted, the threshold boundaries of the diagnostic model are further refined and reset through iterative optimization algorithms to generate optimized threshold boundaries. The above process typically uses gradient descent as the core optimization algorithm. Its basic idea is to minimize the loss function through multiple iterations to obtain the boundary parameters that minimize the model's prediction error. In the initial stage, the threshold is initialized based on the historical model boundaries. Subsequently, the deviation between the current model output and the expected output is calculated using real-time load feature data. The deviation is used as the loss function input into the gradient descent process, and the threshold boundaries are gradually corrected according to the gradient direction, so that the model can still accurately distinguish between normal and abnormal states under new load conditions. During the optimization process, the convergence of the loss is continuously calculated. When the gradient descent reaches the set accuracy range, the final optimized threshold boundaries are output. Through the above methods, the model can maintain the stability and adaptability of boundary determination under load changes, environmental noise, or operating condition disturbances.

[0081] Based on the optimized threshold boundary, the classification rules of the diagnostic model are updated. The diagnostic model then applies the new boundary parameters to the classification function, redefining the judgment logic for the input deviation vector. Specifically, if the magnitude of the deviation vector exceeds the updated boundary, it is marked as an abnormal state; otherwise, it is considered a normal state. This update process is essentially a rule reconstruction of the model's decision layer. By finely adjusting the classification criteria, the model output better reflects the distribution characteristics of the actual operating state. The updated classification rules not only inherit the learning ability of the historical model but also absorb the changing trends of real-time data, thus forming an intelligent judgment mechanism that integrates static empirical knowledge with dynamic operating condition feedback. An optimized diagnostic box is then generated based on the updated classification rules. The optimized diagnostic framework integrates the operating condition classification labels obtained from cluster analysis with the deviation patterns output by the support vector machine model. It achieves closed-loop feedback for fault diagnosis through a hierarchical data structure and a multi-model collaborative mechanism. Specifically, the above diagnostic framework uses the operating condition classification labels as prior information about the operating status and the deviation vector output by the support vector machine as the diagnostic basis. It combines optimized threshold boundaries for multi-dimensional fusion analysis, thereby realizing dynamic judgment and continuous diagnosis of subsequent operating data. The optimized diagnostic framework has self-learning, self-adaptation, and self-calibration capabilities. When the motor operating conditions continuously change, the system can continuously correct the model boundaries through new real-time load characteristic data, achieving long-term adaptation and stable operation in complex working environments.

[0082] The above technical solution achieves dynamic evolution and continuous optimization of the diagnostic model through a complete closed loop of potential fault signal triggering, real-time data acquisition, model parameter updating, threshold boundary optimization, classification rule reconstruction, and diagnostic framework generation. It can maintain high-precision fault identification capability under different loads, working conditions, and environmental disturbances, thereby effectively improving the intelligence level and engineering reliability of the crane motor fault diagnosis system.

[0083] Further, in step S4, the final fault diagnosis result is obtained, including:

[0084] The process involves: acquiring subsequent operational data; extracting features from the subsequent operational data using the optimized diagnostic framework to generate subsequent feature vectors; matching the subsequent feature vectors with the operating condition classification labels to generate classification results; fusing the classification results with the deviation vector to generate a comprehensive diagnostic vector; and comparing the comprehensive diagnostic vector with a preset fault template to obtain the final fault diagnosis result, wherein the final fault diagnosis result characterizes the fault type of the motor.

[0085] Specifically, in order to achieve intelligent identification and accurate classification of the motor under continuous operation through the optimized diagnostic framework, and generate the final fault diagnosis result reflecting the health status of the motor, after the optimized diagnostic framework completes parameter updates and threshold boundary optimization, the subsequent operating data of the motor is collected and analyzed. The subsequent operating data is collected in real time by current sensors, voltage sensors and speed sensors installed on key parts of the motor, and continuously recorded by the data acquisition module to form the subsequent operating sequence. The above sequence fully reflects the energy input and mechanical output characteristics of the motor under dynamic load conditions, providing the original basis for subsequent feature extraction and fault identification.

[0086] After data acquisition, an optimized diagnostic framework is used to extract features from subsequent operational data. First, the acquired multi-channel signals are preprocessed, including frequency domain filtering based on Fourier transform to remove high-frequency noise and electromagnetic interference, thus obtaining a clean operational sequence. The introduction of Fourier transform converts the time-domain signal into a frequency-domain signal, enabling the identification and removal of high-frequency components unrelated to the actual vibration or current fluctuations of the motor, making the filtered signal more accurately reflect the motor's true operating condition. Next, statistical features are extracted from the clean sequence. Parameters such as mean and variance are calculated in the time domain to characterize operational stability, and peak frequency and power spectrum features are extracted in the frequency domain to capture vibration modes and load change trends. These time- and frequency-domain features are combined into a subsequent feature vector. This feature vector is a quantitative expression of the motor's operating state in the feature space, comprehensively reflecting the motor's electrical behavior and mechanical response characteristics, providing input for subsequent classification and judgment. The basic process involves obtaining subsequent feature vectors and then performing matching analysis between these vectors and existing operating condition classification labels to determine the category to which the current operating state belongs. The matching process is implemented using a clustering algorithm, preferably employing the K-means algorithm to calculate the similarity between the subsequent feature vectors and known operating condition templates. The algorithm first initializes cluster centers, calculates the Euclidean distance between the input vectors, and iteratively updates the center positions, ensuring that each vector belongs to the closest cluster according to the minimum distance criterion. If the distance between a subsequent feature vector and a certain operating condition label, such as "high load mode" or "normal operating mode," is below a set threshold, a successful match is achieved, and the corresponding classification result is generated. If the distance exceeds the threshold, the threshold parameter is dynamically adjusted, and the matching is repeated to ensure the accuracy and robustness of the classification results. This classification mechanism enables newly collected data to be quickly assigned to the closest operating condition mode, thereby achieving dynamic identification of operating states and adaptive adjustment of operating condition classification.

[0087] After classification, the classification results are fused with the deviation vector generated by the support vector machine model to form a comprehensive diagnostic vector. The deviation vector represents the degree of deviation of the current operating sequence from the standard healthy sequence, and is an anomaly quantification indicator output by the model layer. The classification results represent the operating mode information of the working condition layer. The classification results are quantified and combined with the corresponding items of each dimension of the deviation vector. A dynamic programming algorithm is used to optimize the weight allocation, ensuring the fusion result maintains balance across different signal dimensions. The dynamic programming algorithm solves the weights through global optimization of the similarity matrix path, ensuring the consistency of the fusion result in terms of time sequence and its interpretability in a physical sense. The fused comprehensive diagnostic vector encompasses multi-source information such as working condition labels, deviation patterns, and signal characteristics, representing a multi-dimensional representation of the motor's current health status. It reflects both the macro-trend of statistical characteristics and preserves... The detailed features of the abnormal components are analyzed. After obtaining the comprehensive diagnostic vector, it is compared with a preset fault template to determine the final fault type. The fault template is a set of feature vectors established based on historical operating data and typical fault samples. Each template represents a known motor fault mode, such as bearing wear, winding short circuit, or rotor imbalance. The cosine similarity between the comprehensive diagnostic vector and each template is calculated to measure their similarity in the feature space. The cosine similarity calculation is based on the normalized ratio of the vector dot product to the modulus, reflecting the angle relationship between the two vectors. The closer the value is to 1, the more similar they are. The template with the highest similarity is selected as the judgment result, and the corresponding fault type is output. For example, when the comprehensive diagnostic vector has the highest similarity to the bearing fault template in terms of high-frequency components and current peak features, the current fault can be determined as "bearing abnormality", and the corresponding diagnostic result and warning signal can be generated.

[0088] Through the above technical solutions, a closed-loop diagnostic process is logically formed, from data acquisition, feature extraction, state classification, information fusion to template matching. This ensures the integrity and continuity of the diagnostic process, enabling the system to maintain high-precision fault identification capabilities under complex loads, multi-source noise, and nonlinear interference conditions. The final fault diagnosis results can not only clearly identify the fault type of the motor, but also quantify the severity of the fault, providing a reliable basis for real-time monitoring, maintenance decisions, and life management of crane motors, thereby realizing intelligent and predictive fault diagnosis of key equipment.

[0089] The present invention also provides a fault diagnosis system for crane motors, for implementing the above-described method, such as... Figure 2 As shown, the system includes:

[0090] The operating condition classification unit is used to acquire crane motor operating data in real time through sensors to obtain an operating sequence; and to obtain an operating condition classification label based on the time-domain statistical features and frequency-domain statistical features extracted from the operating sequence.

[0091] An abnormal deviation calculation unit is used to determine the operating mode of the crane motor based on the operating condition classification label, determine the deviation vector between the operating sequence and the standard healthy sequence through a support vector machine model, and calculate the abnormal deviation threshold.

[0092] The similarity analysis unit is used to obtain the similarity matrix between the current running sequence and the historical sequence based on the abnormal deviation threshold, adjust the similarity matrix through a dynamic programming algorithm, and obtain the weight set of the adjusted similarity matrix.

[0093] The fault diagnosis unit is used to fuse operational data through an adjusted weight set to form a comprehensive diagnostic index. Based on the comprehensive diagnostic index, it obtains a comprehensive diagnostic index, determines whether the comprehensive diagnostic index exceeds a threshold, and obtains potential fault signals. It updates the diagnostic model parameters according to the potential fault signals to obtain an optimized diagnostic framework. Based on the optimized diagnostic framework, it analyzes operational data, fuses the operating condition classification label and the deviation vector, and obtains the final fault diagnosis result.

[0094] The present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0095] In summary, this invention uses sensors to collect current, voltage, and speed signals during motor operation in real time, generating an operating sequence. After filtering, normalization, and time-frequency domain feature extraction, a high-quality data foundation is provided for subsequent analysis. This sequence reflects both the dynamic response of the motor and retains load fluctuation characteristics, providing rich information support for operating condition classification. A clustering model is used to group and analyze the feature vectors of the operating sequence, and operating condition classification labels are generated by combining them with preset operating condition templates, enabling automatic identification of the motor's operating mode and providing accurate status indicators for subsequent deviation calculations. Based on the operating condition identification results, a support vector machine model is used to calculate the deviation vector of the operating sequence relative to a standard healthy sequence, and an abnormal deviation threshold is determined accordingly. Through threshold verification and boundary optimization, it can adapt to different noise environments and load conditions, thereby obtaining a more stable anomaly detection capability. Based on this threshold, a reference sequence is selected from historical operating sequences, and dynamic regulation is applied... The algorithm calculates the similarity between the current sequence and the reference sequence, and optimizes the weight set to achieve the fusion of multi-dimensional data. The fused signal is further used to extract comprehensive features, and the model calculates comprehensive diagnostic indicators to determine whether there are potential fault signals. When a potential anomaly is detected, the model parameters are updated using real-time load feature data, and the classification boundary is adjusted through an iterative optimization algorithm to generate an optimized diagnostic framework. The optimized framework performs feature analysis and classification matching on newly collected operating data, combines the deviation vector to form a comprehensive diagnostic vector, and compares it with a preset fault template to output the final fault type. Through the cooperation of the above technical solutions, dynamic diagnosis and adaptive learning of crane motors under multiple working conditions and multiple signal environments are realized. It can not only accurately identify early anomalies, but also optimize the model in real time according to the operating status, realizing a closed-loop diagnosis from data acquisition to intelligent decision-making, which significantly improves the accuracy, real-time performance and robustness of fault identification.

[0096] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0097] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault diagnosis method for crane motors, characterized in that, The method includes: Step S1: Acquire crane motor operation data in real time through sensors to obtain an operation sequence; obtain operating condition classification labels based on the time-domain statistical features and frequency-domain statistical features extracted from the operation sequence; Step S2: Determine the crane motor's operating mode based on the operating condition classification label; determine the deviation vector between the operating sequence and the standard healthy sequence using a support vector machine model; calculate the abnormal deviation threshold; obtain the similarity matrix between the current operating sequence and historical sequences based on the abnormal deviation threshold; adjust the similarity matrix using a dynamic programming algorithm; and obtain the weight set of the adjusted similarity matrix. Obtaining the abnormal deviation threshold includes: if the operating condition classification label indicates a high-load mode, activating the pre-trained support vector machine model; analyzing the operating sequence using the support vector machine model to generate a deviation vector, where the deviation vector represents the degree of deviation in the operating sequence; calculating the abnormal deviation threshold based on the distribution characteristics of the deviation vector; verifying the abnormal deviation threshold and adjusting the classification boundary; updating the abnormal deviation threshold based on the adjusted classification boundary, where the abnormal deviation threshold is used to determine abnormal operation. Obtaining the adjusted weight set includes: acquiring a historical running sequence dataset; selecting reference sequences from the historical running sequence dataset based on the abnormal deviation threshold; calculating the similarity between the current running sequence and the reference sequences to generate a similarity matrix; optimizing the weight coefficients of the similarity matrix using a dynamic programming algorithm to obtain an initial weight set; and normalizing the initial weight set to generate the adjusted weight set. Step S3: By integrating the operational data through the adjusted weight set, a comprehensive diagnostic index is formed. It is then determined whether the comprehensive diagnostic index exceeds the threshold to obtain potential fault signals. Step S4: Update the diagnostic model parameters based on the potential fault signals to obtain an optimized diagnostic framework; analyze the operating data based on the optimized diagnostic framework, fuse the operating condition classification label and the deviation vector to obtain the final fault diagnosis result.

2. The method as described in claim 1, characterized in that, In step S1, the running sequence is obtained, including: The current, voltage, and speed signals of the motor are acquired by sensors; the current, voltage, and speed signals are processed to filter out noise and obtain an operating sequence; the operating sequence is analyzed in the time domain to extract time series features; and the operating sequence is analyzed in the frequency domain to extract frequency distribution features.

3. The method as described in claim 2, characterized in that, In step S1, the working condition classification labels are obtained, including: The time-domain statistical features and frequency-domain statistical features of the running sequence are combined into a feature vector; the feature vector is grouped and analyzed by a pre-trained clustering model to obtain an initial classification result, wherein the time-domain statistical features include mean, variance and peak value, and the frequency-domain statistical features include power spectral density and dominant frequency component; The initial classification result is matched with a preset operating condition template to generate the operating condition classification label, wherein the operating condition classification label represents the operating status of the motor.

4. The method as described in claim 1, characterized in that, In step S3, potential fault signals are obtained, including: Acquire the current signal and speed signal from the running sequence; perform weighted fusion of the current signal and speed signal according to the adjusted weight set to generate a fused signal; perform feature extraction on the fused signal to obtain a comprehensive feature vector; A comprehensive diagnostic index is calculated based on the comprehensive feature vector, wherein the comprehensive diagnostic index represents the operating health status of the motor; it is determined whether the comprehensive diagnostic index exceeds a preset diagnostic threshold, and a potential fault signal is generated based on the determination result.

5. The method as described in claim 1, characterized in that, In step S4, the optimized diagnostic framework is obtained, including: If the potential fault signal exists, real-time load feature data is acquired; the parameters of the pre-trained diagnostic model are updated based on the real-time load feature data; the threshold boundary of the diagnostic model is adjusted through an iterative optimization algorithm to generate an optimized threshold boundary; the classification rules of the diagnostic model are updated based on the optimized threshold boundary; and an optimized diagnostic framework is generated based on the updated classification rules.

6. The method as described in claim 1, characterized in that, In step S4, the final fault diagnosis result is obtained, including: The process involves: acquiring subsequent operational data; extracting features from the subsequent operational data using the optimized diagnostic framework to generate subsequent feature vectors; matching the subsequent feature vectors with the operating condition classification labels to generate classification results; fusing the classification results with the deviation vector to generate a comprehensive diagnostic vector; and comparing the comprehensive diagnostic vector with a preset fault template to obtain the final fault diagnosis result, wherein the final fault diagnosis result characterizes the fault type of the motor.

7. A fault diagnosis system for a crane motor, used to implement the method as described in any one of claims 1-6, characterized in that, The system includes: The operating condition classification unit is used to acquire crane motor operating data in real time through sensors to obtain an operating sequence; and to obtain an operating condition classification label based on the time-domain statistical features and frequency-domain statistical features extracted from the operating sequence. An abnormal deviation calculation unit is used to determine the operating mode of the crane motor based on the operating condition classification label, determine the deviation vector between the operating sequence and the standard healthy sequence through a support vector machine model, and calculate the abnormal deviation threshold. The similarity analysis unit is used to obtain the similarity matrix between the current running sequence and the historical sequence based on the abnormal deviation threshold, adjust the similarity matrix through a dynamic programming algorithm, and obtain the weight set of the adjusted similarity matrix. The fault diagnosis unit is used to fuse operational data through an adjusted weight set to form a comprehensive diagnostic index, determine whether the comprehensive diagnostic index exceeds a threshold, and obtain potential fault signals; update the diagnostic model parameters based on the potential fault signals to obtain an optimized diagnostic framework; analyze operational data based on the optimized diagnostic framework, fuse the operating condition classification label and the deviation vector, and obtain the final fault diagnosis result.

8. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method as described in any one of claims 1-6.

Images